< 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 "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.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 "interpreter/interpreterRuntime.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 "prims/methodHandles.hpp"

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

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




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

 1286 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1287   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1288   assert_different_registers(arg_1, c_rarg2);
 1289   pass_arg2(this, arg_2);
 1290   pass_arg1(this, arg_1);
 1291   pass_arg0(this, arg_0);
 1292   call_VM_leaf(entry_point, 3);
 1293 }
 1294 
 1295 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1296   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1297   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1298   assert_different_registers(arg_2, c_rarg3);
 1299   pass_arg3(this, arg_3);
 1300   pass_arg2(this, arg_2);
 1301   pass_arg1(this, arg_1);
 1302   pass_arg0(this, arg_0);
 1303   call_VM_leaf(entry_point, 3);
 1304 }
 1305 




 1306 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1307   pass_arg0(this, arg_0);
 1308   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1309 }
 1310 
 1311 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1312   assert_different_registers(arg_0, c_rarg1);
 1313   pass_arg1(this, arg_1);
 1314   pass_arg0(this, arg_0);
 1315   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1316 }
 1317 
 1318 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1319   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1320   assert_different_registers(arg_1, c_rarg2);
 1321   pass_arg2(this, arg_2);
 1322   pass_arg1(this, arg_1);
 1323   pass_arg0(this, arg_0);
 1324   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1325 }

 2339     lea(rscratch, src);
 2340     Assembler::mulss(dst, Address(rscratch, 0));
 2341   }
 2342 }
 2343 
 2344 void MacroAssembler::null_check(Register reg, int offset) {
 2345   if (needs_explicit_null_check(offset)) {
 2346     // provoke OS null exception if reg is null by
 2347     // accessing M[reg] w/o changing any (non-CC) registers
 2348     // NOTE: cmpl is plenty here to provoke a segv
 2349     cmpptr(rax, Address(reg, 0));
 2350     // Note: should probably use testl(rax, Address(reg, 0));
 2351     //       may be shorter code (however, this version of
 2352     //       testl needs to be implemented first)
 2353   } else {
 2354     // nothing to do, (later) access of M[reg + offset]
 2355     // will provoke OS null exception if reg is null
 2356   }
 2357 }
 2358 











































































































 2359 void MacroAssembler::os_breakpoint() {
 2360   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2361   // (e.g., MSVC can't call ps() otherwise)
 2362   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2363 }
 2364 
 2365 void MacroAssembler::unimplemented(const char* what) {
 2366   const char* buf = nullptr;
 2367   {
 2368     ResourceMark rm;
 2369     stringStream ss;
 2370     ss.print("unimplemented: %s", what);
 2371     buf = code_string(ss.as_string());
 2372   }
 2373   stop(buf);
 2374 }
 2375 
 2376 #define XSTATE_BV 0x200
 2377 
 2378 void MacroAssembler::pop_CPU_state() {

 3443 }
 3444 
 3445 // C++ bool manipulation
 3446 void MacroAssembler::testbool(Register dst) {
 3447   if(sizeof(bool) == 1)
 3448     testb(dst, 0xff);
 3449   else if(sizeof(bool) == 2) {
 3450     // testw implementation needed for two byte bools
 3451     ShouldNotReachHere();
 3452   } else if(sizeof(bool) == 4)
 3453     testl(dst, dst);
 3454   else
 3455     // unsupported
 3456     ShouldNotReachHere();
 3457 }
 3458 
 3459 void MacroAssembler::testptr(Register dst, Register src) {
 3460   testq(dst, src);
 3461 }
 3462 






















































































































 3463 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3464 void MacroAssembler::tlab_allocate(Register obj,
 3465                                    Register var_size_in_bytes,
 3466                                    int con_size_in_bytes,
 3467                                    Register t1,
 3468                                    Register t2,
 3469                                    Label& slow_case) {
 3470   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3471   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3472 }
 3473 
 3474 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3475   RegSet regs;
 3476   regs += RegSet::of(rax, rcx, rdx);
 3477 #ifndef _WINDOWS
 3478   regs += RegSet::of(rsi, rdi);
 3479 #endif
 3480   regs += RegSet::range(r8, r11);
 3481   if (UseAPX) {
 3482     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));

 3646   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3647   if (UseIncDec) {
 3648     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3649   } else {
 3650     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3651     shrptr(index, 1);
 3652   }
 3653 
 3654   // initialize remaining object fields: index is a multiple of 2 now
 3655   {
 3656     Label loop;
 3657     bind(loop);
 3658     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3659     decrement(index);
 3660     jcc(Assembler::notZero, loop);
 3661   }
 3662 
 3663   bind(done);
 3664 }
 3665 






















 3666 // Look up the method for a megamorphic invokeinterface call.
 3667 // The target method is determined by <intf_klass, itable_index>.
 3668 // The receiver klass is in recv_klass.
 3669 // On success, the result will be in method_result, and execution falls through.
 3670 // On failure, execution transfers to the given label.
 3671 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3672                                              Register intf_klass,
 3673                                              RegisterOrConstant itable_index,
 3674                                              Register method_result,
 3675                                              Register scan_temp,
 3676                                              Label& L_no_such_interface,
 3677                                              bool return_method) {
 3678   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3679   assert_different_registers(method_result, intf_klass, scan_temp);
 3680   assert(recv_klass != method_result || !return_method,
 3681          "recv_klass can be destroyed when method isn't needed");
 3682 
 3683   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3684          "caller must use same register for non-constant itable index as for method");
 3685 

 4696   } else {
 4697     Label L;
 4698     jccb(negate_condition(cc), L);
 4699     movl(dst, src);
 4700     bind(L);
 4701   }
 4702 }
 4703 
 4704 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4705   if (VM_Version::supports_cmov()) {
 4706     cmovl(cc, dst, src);
 4707   } else {
 4708     Label L;
 4709     jccb(negate_condition(cc), L);
 4710     movl(dst, src);
 4711     bind(L);
 4712   }
 4713 }
 4714 
 4715 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4716   if (!VerifyOops) return;




 4717 
 4718   BLOCK_COMMENT("verify_oop {");
 4719   push(rscratch1);
 4720   push(rax);                          // save rax
 4721   push(reg);                          // pass register argument
 4722 
 4723   // Pass register number to verify_oop_subroutine
 4724   const char* b = nullptr;
 4725   {
 4726     ResourceMark rm;
 4727     stringStream ss;
 4728     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4729     b = code_string(ss.as_string());
 4730   }
 4731   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4732   pushptr(buffer.addr(), rscratch1);
 4733 
 4734   // call indirectly to solve generation ordering problem
 4735   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4736   call(rax);

 4952   // or something else. Since this is a slow path, we can optimize for code density,
 4953   // and just restart the search from the beginning.
 4954   jmpb(L_restart);
 4955 
 4956   // Counter updates:
 4957 
 4958   // Increment polymorphic counter instead of receiver slot.
 4959   bind(L_polymorphic);
 4960   movptr(offset, poly_count_offset);
 4961   jmpb(L_count_update);
 4962 
 4963   // Found a receiver, convert its slot offset to corresponding count offset.
 4964   bind(L_found_recv);
 4965   addptr(offset, receiver_to_count_step);
 4966 
 4967   bind(L_count_update);
 4968   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 4969 }
 4970 
 4971 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 4972   if (!VerifyOops) return;




 4973 
 4974   push(rscratch1);
 4975   push(rax); // save rax,
 4976   // addr may contain rsp so we will have to adjust it based on the push
 4977   // we just did (and on 64 bit we do two pushes)
 4978   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 4979   // stores rax into addr which is backwards of what was intended.
 4980   if (addr.uses(rsp)) {
 4981     lea(rax, addr);
 4982     pushptr(Address(rax, 2 * BytesPerWord));
 4983   } else {
 4984     pushptr(addr);
 4985   }
 4986 
 4987   // Pass register number to verify_oop_subroutine
 4988   const char* b = nullptr;
 4989   {
 4990     ResourceMark rm;
 4991     stringStream ss;
 4992     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);

 5346 
 5347 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5348   // get mirror
 5349   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5350   load_method_holder(mirror, method);
 5351   movptr(mirror, Address(mirror, mirror_offset));
 5352   resolve_oop_handle(mirror, tmp);
 5353 }
 5354 
 5355 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5356   load_method_holder(rresult, rmethod);
 5357   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5358 }
 5359 
 5360 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5361   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5362   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5363   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5364 }
 5365 










 5366 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5367   assert(UseCompactObjectHeaders, "expect compact object headers");
 5368   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5369   shrq(dst, markWord::klass_shift);
 5370 }
 5371 
 5372 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5373   assert_different_registers(src, tmp);
 5374   assert_different_registers(dst, tmp);
 5375 
 5376   if (UseCompactObjectHeaders) {
 5377     load_narrow_klass_compact(dst, src);
 5378     decode_klass_not_null(dst, tmp);
 5379   } else if (UseCompressedClassPointers) {
 5380     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5381     decode_klass_not_null(dst, tmp);
 5382   } else {
 5383     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5384   }
 5385 }
 5386 





 5387 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5388   assert(!UseCompactObjectHeaders, "not with compact headers");
 5389   assert_different_registers(src, tmp);
 5390   assert_different_registers(dst, tmp);
 5391   if (UseCompressedClassPointers) {
 5392     encode_klass_not_null(src, tmp);
 5393     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5394   } else {
 5395     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5396   }
 5397 }
 5398 
 5399 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5400   if (UseCompactObjectHeaders) {
 5401     assert(tmp != noreg, "need tmp");
 5402     assert_different_registers(klass, obj, tmp);
 5403     load_narrow_klass_compact(tmp, obj);
 5404     cmpl(klass, tmp);
 5405   } else if (UseCompressedClassPointers) {
 5406     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));

 5432   bool as_raw = (decorators & AS_RAW) != 0;
 5433   if (as_raw) {
 5434     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5435   } else {
 5436     bs->load_at(this, decorators, type, dst, src, tmp1);
 5437   }
 5438 }
 5439 
 5440 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5441                                      Register tmp1, Register tmp2, Register tmp3) {
 5442   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5443   decorators = AccessInternal::decorator_fixup(decorators, type);
 5444   bool as_raw = (decorators & AS_RAW) != 0;
 5445   if (as_raw) {
 5446     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5447   } else {
 5448     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5449   }
 5450 }
 5451 








































 5452 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5453   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5454 }
 5455 
 5456 // Doesn't do verification, generates fixed size code
 5457 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5458   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5459 }
 5460 
 5461 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5462                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5463   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5464 }
 5465 
 5466 // Used for storing nulls.
 5467 void MacroAssembler::store_heap_oop_null(Address dst) {
 5468   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5469 }
 5470 
 5471 void MacroAssembler::store_klass_gap(Register dst, Register src) {

 5788   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 5789   int klass_index = oop_recorder()->find_index(k);
 5790   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 5791   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 5792 }
 5793 
 5794 void MacroAssembler::reinit_heapbase() {
 5795   if (UseCompressedOops) {
 5796     if (Universe::heap() != nullptr) {
 5797       if (CompressedOops::base() == nullptr) {
 5798         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 5799       } else {
 5800         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 5801       }
 5802     } else {
 5803       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 5804     }
 5805   }
 5806 }
 5807 
















































































































































































































































































































































































































































































 5808 #if COMPILER2_OR_JVMCI
 5809 
 5810 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 5811 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5812   // cnt - number of qwords (8-byte words).
 5813   // base - start address, qword aligned.
 5814   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 5815   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 5816   if (use64byteVector) {
 5817     vpxor(xtmp, xtmp, xtmp, AVX_512bit);
 5818   } else if (MaxVectorSize >= 32) {
 5819     vpxor(xtmp, xtmp, xtmp, AVX_256bit);


 5820   } else {
 5821     pxor(xtmp, xtmp);

 5822   }
 5823   jmp(L_zero_64_bytes);
 5824 
 5825   BIND(L_loop);
 5826   if (MaxVectorSize >= 32) {
 5827     fill64(base, 0, xtmp, use64byteVector);
 5828   } else {
 5829     movdqu(Address(base,  0), xtmp);
 5830     movdqu(Address(base, 16), xtmp);
 5831     movdqu(Address(base, 32), xtmp);
 5832     movdqu(Address(base, 48), xtmp);
 5833   }
 5834   addptr(base, 64);
 5835 
 5836   BIND(L_zero_64_bytes);
 5837   subptr(cnt, 8);
 5838   jccb(Assembler::greaterEqual, L_loop);
 5839 
 5840   // Copy trailing 64 bytes
 5841   if (use64byteVector) {
 5842     addptr(cnt, 8);
 5843     jccb(Assembler::equal, L_end);
 5844     fill64_masked(3, base, 0, xtmp, mask, cnt, rtmp, true);
 5845     jmp(L_end);
 5846   } else {
 5847     addptr(cnt, 4);
 5848     jccb(Assembler::less, L_tail);
 5849     if (MaxVectorSize >= 32) {
 5850       vmovdqu(Address(base, 0), xtmp);
 5851     } else {
 5852       movdqu(Address(base,  0), xtmp);
 5853       movdqu(Address(base, 16), xtmp);
 5854     }
 5855   }
 5856   addptr(base, 32);
 5857   subptr(cnt, 4);
 5858 
 5859   BIND(L_tail);
 5860   addptr(cnt, 4);
 5861   jccb(Assembler::lessEqual, L_end);
 5862   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 5863     fill32_masked(3, base, 0, xtmp, mask, cnt, rtmp);
 5864   } else {
 5865     decrement(cnt);
 5866 
 5867     BIND(L_sloop);
 5868     movq(Address(base, 0), xtmp);
 5869     addptr(base, 8);
 5870     decrement(cnt);
 5871     jccb(Assembler::greaterEqual, L_sloop);
 5872   }
 5873   BIND(L_end);
 5874 }
 5875 
 5876 // Clearing constant sized memory using YMM/ZMM registers.
 5877 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 5878   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 5879   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 5880 
 5881   int vector64_count = (cnt & (~0x7)) >> 3;
 5882   cnt = cnt & 0x7;
 5883   const int fill64_per_loop = 4;

 5945         break;
 5946       case 7:
 5947         if (use64byteVector) {
 5948           movl(rtmp, 0x7F);
 5949           kmovwl(mask, rtmp);
 5950           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 5951         } else {
 5952           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 5953           movl(rtmp, 0x7);
 5954           kmovwl(mask, rtmp);
 5955           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 5956         }
 5957         break;
 5958       default:
 5959         fatal("Unexpected length : %d\n",cnt);
 5960         break;
 5961     }
 5962   }
 5963 }
 5964 
 5965 void MacroAssembler::clear_mem(Register base, Register cnt, Register tmp, XMMRegister xtmp,
 5966                                bool is_large, KRegister mask) {
 5967   // cnt      - number of qwords (8-byte words).
 5968   // base     - start address, qword aligned.
 5969   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 5970   assert(base==rdi, "base register must be edi for rep stos");
 5971   assert(tmp==rax,   "tmp register must be eax for rep stos");
 5972   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 5973   assert(InitArrayShortSize % BytesPerLong == 0,
 5974     "InitArrayShortSize should be the multiple of BytesPerLong");
 5975 
 5976   Label DONE;
 5977   if (!is_large || !UseXMMForObjInit) {
 5978     xorptr(tmp, tmp);
 5979   }
 5980 
 5981   if (!is_large) {
 5982     Label LOOP, LONG;
 5983     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 5984     jccb(Assembler::greater, LONG);
 5985 
 5986     decrement(cnt);
 5987     jccb(Assembler::negative, DONE); // Zero length
 5988 
 5989     // Use individual pointer-sized stores for small counts:
 5990     BIND(LOOP);
 5991     movptr(Address(base, cnt, Address::times_ptr), tmp);
 5992     decrement(cnt);
 5993     jccb(Assembler::greaterEqual, LOOP);
 5994     jmpb(DONE);
 5995 
 5996     BIND(LONG);
 5997   }
 5998 
 5999   // Use longer rep-prefixed ops for non-small counts:
 6000   if (UseFastStosb) {
 6001     shlptr(cnt, 3); // convert to number of bytes
 6002     rep_stosb();
 6003   } else if (UseXMMForObjInit) {
 6004     xmm_clear_mem(base, cnt, tmp, xtmp, mask);
 6005   } else {
 6006     rep_stos();
 6007   }
 6008 
 6009   BIND(DONE);
 6010 }
 6011 
 6012 #endif //COMPILER2_OR_JVMCI
 6013 
 6014 
 6015 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6016                                    Register to, Register value, Register count,
 6017                                    Register rtmp, XMMRegister xtmp) {
 6018   ShortBranchVerifier sbv(this);
 6019   assert_different_registers(to, value, count, rtmp);
 6020   Label L_exit;
 6021   Label L_fill_2_bytes, L_fill_4_bytes;
 6022 
 6023 #if defined(COMPILER2)
 6024   if(MaxVectorSize >=32 &&

 9904 
 9905   // Load top.
 9906   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9907 
 9908   // Check if the lock-stack is full.
 9909   cmpl(top, LockStack::end_offset());
 9910   jcc(Assembler::greaterEqual, slow);
 9911 
 9912   // Check for recursion.
 9913   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
 9914   jcc(Assembler::equal, push);
 9915 
 9916   // Check header for monitor (0b10).
 9917   testptr(reg_rax, markWord::monitor_value);
 9918   jcc(Assembler::notZero, slow);
 9919 
 9920   // Try to lock. Transition lock bits 0b01 => 0b00
 9921   movptr(tmp, reg_rax);
 9922   andptr(tmp, ~(int32_t)markWord::unlocked_value);
 9923   orptr(reg_rax, markWord::unlocked_value);



 9924   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
 9925   jcc(Assembler::notEqual, slow);
 9926 
 9927   // Restore top, CAS clobbers register.
 9928   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
 9929 
 9930   bind(push);
 9931   // After successful lock, push object on lock-stack.
 9932   movptr(Address(thread, top), obj);
 9933   incrementl(top, oopSize);
 9934   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
 9935 }
 9936 
 9937 // Implements fast-unlocking.
 9938 //
 9939 // obj: the object to be unlocked
 9940 // reg_rax: rax
 9941 // thread: the thread
 9942 // tmp: a temporary register
 9943 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, 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 "asm/assembler.hpp"
   26 #include "asm/assembler.inline.hpp"
   27 #include "code/aotCodeCache.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 "interpreter/interpreterRuntime.hpp"
   40 #include "jvm.h"
   41 #include "memory/resourceArea.hpp"
   42 #include "memory/universe.hpp"
   43 #include "oops/accessDecorators.hpp"
   44 #include "oops/compressedKlass.inline.hpp"
   45 #include "oops/compressedOops.inline.hpp"
   46 #include "oops/klass.inline.hpp"
   47 #include "oops/resolvedFieldEntry.hpp"
   48 #include "prims/methodHandles.hpp"
   49 #include "runtime/arguments.hpp"
   50 #include "runtime/continuation.hpp"
   51 #include "runtime/interfaceSupport.inline.hpp"
   52 #include "runtime/javaThread.hpp"
   53 #include "runtime/jniHandles.hpp"
   54 #include "runtime/objectMonitor.hpp"
   55 #include "runtime/os.hpp"
   56 #include "runtime/safepoint.hpp"
   57 #include "runtime/safepointMechanism.hpp"
   58 #include "runtime/sharedRuntime.hpp"
   59 #include "runtime/signature_cc.hpp"
   60 #include "runtime/stubRoutines.hpp"
   61 #include "utilities/checkedCast.hpp"
   62 #include "utilities/macros.hpp"
   63 #include "vmreg_x86.inline.hpp"
   64 #ifdef COMPILER2
   65 #include "opto/output.hpp"
   66 #endif
   67 
   68 #ifdef PRODUCT
   69 #define BLOCK_COMMENT(str) /* nothing */
   70 #define STOP(error) stop(error)
   71 #else
   72 #define BLOCK_COMMENT(str) block_comment(str)
   73 #define STOP(error) block_comment(error); stop(error)
   74 #endif
   75 
   76 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
   77 
   78 #ifdef ASSERT
   79 bool AbstractAssembler::pd_check_instruction_mark() { return true; }
   80 #endif
   81 
   82 static const Assembler::Condition reverse[] = {
   83     Assembler::noOverflow     /* overflow      = 0x0 */ ,
   84     Assembler::overflow       /* noOverflow    = 0x1 */ ,
   85     Assembler::aboveEqual     /* carrySet      = 0x2, below         = 0x2 */ ,
   86     Assembler::below          /* aboveEqual    = 0x3, carryClear    = 0x3 */ ,

 1294 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1295   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1296   assert_different_registers(arg_1, c_rarg2);
 1297   pass_arg2(this, arg_2);
 1298   pass_arg1(this, arg_1);
 1299   pass_arg0(this, arg_0);
 1300   call_VM_leaf(entry_point, 3);
 1301 }
 1302 
 1303 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2, Register arg_3) {
 1304   assert_different_registers(arg_0, c_rarg1, c_rarg2, c_rarg3);
 1305   assert_different_registers(arg_1, c_rarg2, c_rarg3);
 1306   assert_different_registers(arg_2, c_rarg3);
 1307   pass_arg3(this, arg_3);
 1308   pass_arg2(this, arg_2);
 1309   pass_arg1(this, arg_1);
 1310   pass_arg0(this, arg_0);
 1311   call_VM_leaf(entry_point, 3);
 1312 }
 1313 
 1314 void MacroAssembler::super_call_VM_leaf(address entry_point) {
 1315   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1316 }
 1317 
 1318 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0) {
 1319   pass_arg0(this, arg_0);
 1320   MacroAssembler::call_VM_leaf_base(entry_point, 1);
 1321 }
 1322 
 1323 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1) {
 1324   assert_different_registers(arg_0, c_rarg1);
 1325   pass_arg1(this, arg_1);
 1326   pass_arg0(this, arg_0);
 1327   MacroAssembler::call_VM_leaf_base(entry_point, 2);
 1328 }
 1329 
 1330 void MacroAssembler::super_call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) {
 1331   assert_different_registers(arg_0, c_rarg1, c_rarg2);
 1332   assert_different_registers(arg_1, c_rarg2);
 1333   pass_arg2(this, arg_2);
 1334   pass_arg1(this, arg_1);
 1335   pass_arg0(this, arg_0);
 1336   MacroAssembler::call_VM_leaf_base(entry_point, 3);
 1337 }

 2351     lea(rscratch, src);
 2352     Assembler::mulss(dst, Address(rscratch, 0));
 2353   }
 2354 }
 2355 
 2356 void MacroAssembler::null_check(Register reg, int offset) {
 2357   if (needs_explicit_null_check(offset)) {
 2358     // provoke OS null exception if reg is null by
 2359     // accessing M[reg] w/o changing any (non-CC) registers
 2360     // NOTE: cmpl is plenty here to provoke a segv
 2361     cmpptr(rax, Address(reg, 0));
 2362     // Note: should probably use testl(rax, Address(reg, 0));
 2363     //       may be shorter code (however, this version of
 2364     //       testl needs to be implemented first)
 2365   } else {
 2366     // nothing to do, (later) access of M[reg + offset]
 2367     // will provoke OS null exception if reg is null
 2368   }
 2369 }
 2370 
 2371 void MacroAssembler::test_markword_is_inline_type(Register markword, Label& is_inline_type) {
 2372   andptr(markword, markWord::inline_type_mask_in_place);
 2373   cmpptr(markword, markWord::inline_type_pattern);
 2374   jcc(Assembler::equal, is_inline_type);
 2375 }
 2376 
 2377 void MacroAssembler::test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null) {
 2378   if (can_be_null) {
 2379     testptr(object, object);
 2380     jcc(Assembler::zero, not_inline_type);
 2381   }
 2382   const int is_inline_type_mask = markWord::inline_type_pattern;
 2383   movptr(tmp, Address(object, oopDesc::mark_offset_in_bytes()));
 2384   andptr(tmp, is_inline_type_mask);
 2385   cmpptr(tmp, is_inline_type_mask);
 2386   jcc(Assembler::notEqual, not_inline_type);
 2387 }
 2388 
 2389 void MacroAssembler::test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free_inline_type) {
 2390   movl(temp_reg, flags);
 2391   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2392   jcc(Assembler::notEqual, is_null_free_inline_type);
 2393 }
 2394 
 2395 void MacroAssembler::test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free_inline_type) {
 2396   movl(temp_reg, flags);
 2397   testl(temp_reg, 1 << ResolvedFieldEntry::is_null_free_inline_type_shift);
 2398   jcc(Assembler::equal, not_null_free_inline_type);
 2399 }
 2400 
 2401 void MacroAssembler::test_field_is_flat(Register flags, Register temp_reg, Label& is_flat) {
 2402   movl(temp_reg, flags);
 2403   testl(temp_reg, 1 << ResolvedFieldEntry::is_flat_shift);
 2404   jcc(Assembler::notEqual, is_flat);
 2405 }
 2406 
 2407 void MacroAssembler::test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker) {
 2408   movl(temp_reg, flags);
 2409   testl(temp_reg, 1 << ResolvedFieldEntry::has_null_marker_shift);
 2410   jcc(Assembler::notEqual, has_null_marker);
 2411 }
 2412 
 2413 void MacroAssembler::test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label) {
 2414   Label test_mark_word;
 2415   // load mark word
 2416   movptr(temp_reg, Address(oop, oopDesc::mark_offset_in_bytes()));
 2417   // check displaced
 2418   testl(temp_reg, markWord::unlocked_value);
 2419   jccb(Assembler::notZero, test_mark_word);
 2420   // slow path use klass prototype
 2421   push(rscratch1);
 2422   load_prototype_header(temp_reg, oop, rscratch1);
 2423   pop(rscratch1);
 2424 
 2425   bind(test_mark_word);
 2426   testl(temp_reg, test_bit);
 2427   jcc((jmp_set) ? Assembler::notZero : Assembler::zero, jmp_label);
 2428 }
 2429 
 2430 void MacroAssembler::test_flat_array_oop(Register oop, Register temp_reg,
 2431                                          Label& is_flat_array) {
 2432 #ifdef _LP64
 2433   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, true, is_flat_array);
 2434 #else
 2435   load_klass(temp_reg, oop, noreg);
 2436   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2437   test_flat_array_layout(temp_reg, is_flat_array);
 2438 #endif
 2439 }
 2440 
 2441 void MacroAssembler::test_non_flat_array_oop(Register oop, Register temp_reg,
 2442                                              Label& is_non_flat_array) {
 2443 #ifdef _LP64
 2444   test_oop_prototype_bit(oop, temp_reg, markWord::flat_array_bit_in_place, false, is_non_flat_array);
 2445 #else
 2446   load_klass(temp_reg, oop, noreg);
 2447   movl(temp_reg, Address(temp_reg, Klass::layout_helper_offset()));
 2448   test_non_flat_array_layout(temp_reg, is_non_flat_array);
 2449 #endif
 2450 }
 2451 
 2452 void MacroAssembler::test_null_free_array_oop(Register oop, Register temp_reg, Label&is_null_free_array) {
 2453 #ifdef _LP64
 2454   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, true, is_null_free_array);
 2455 #else
 2456   Unimplemented();
 2457 #endif
 2458 }
 2459 
 2460 void MacroAssembler::test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array) {
 2461 #ifdef _LP64
 2462   test_oop_prototype_bit(oop, temp_reg, markWord::null_free_array_bit_in_place, false, is_non_null_free_array);
 2463 #else
 2464   Unimplemented();
 2465 #endif
 2466 }
 2467 
 2468 void MacroAssembler::test_flat_array_layout(Register lh, Label& is_flat_array) {
 2469   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2470   jcc(Assembler::notZero, is_flat_array);
 2471 }
 2472 
 2473 void MacroAssembler::test_non_flat_array_layout(Register lh, Label& is_non_flat_array) {
 2474   testl(lh, Klass::_lh_array_tag_flat_value_bit_inplace);
 2475   jcc(Assembler::zero, is_non_flat_array);
 2476 }
 2477 
 2478 void MacroAssembler::os_breakpoint() {
 2479   // instead of directly emitting a breakpoint, call os:breakpoint for better debugability
 2480   // (e.g., MSVC can't call ps() otherwise)
 2481   call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint)));
 2482 }
 2483 
 2484 void MacroAssembler::unimplemented(const char* what) {
 2485   const char* buf = nullptr;
 2486   {
 2487     ResourceMark rm;
 2488     stringStream ss;
 2489     ss.print("unimplemented: %s", what);
 2490     buf = code_string(ss.as_string());
 2491   }
 2492   stop(buf);
 2493 }
 2494 
 2495 #define XSTATE_BV 0x200
 2496 
 2497 void MacroAssembler::pop_CPU_state() {

 3562 }
 3563 
 3564 // C++ bool manipulation
 3565 void MacroAssembler::testbool(Register dst) {
 3566   if(sizeof(bool) == 1)
 3567     testb(dst, 0xff);
 3568   else if(sizeof(bool) == 2) {
 3569     // testw implementation needed for two byte bools
 3570     ShouldNotReachHere();
 3571   } else if(sizeof(bool) == 4)
 3572     testl(dst, dst);
 3573   else
 3574     // unsupported
 3575     ShouldNotReachHere();
 3576 }
 3577 
 3578 void MacroAssembler::testptr(Register dst, Register src) {
 3579   testq(dst, src);
 3580 }
 3581 
 3582 // Object / value buffer allocation...
 3583 //
 3584 // Kills klass and rsi on LP64
 3585 void MacroAssembler::allocate_instance(Register klass, Register new_obj,
 3586                                        Register t1, Register t2,
 3587                                        bool clear_fields, Label& alloc_failed)
 3588 {
 3589   Label done, initialize_header, initialize_object, slow_case, slow_case_no_pop;
 3590   Register layout_size = t1;
 3591   assert(new_obj == rax, "needs to be rax");
 3592   assert_different_registers(klass, new_obj, t1, t2);
 3593 
 3594   // get instance_size in InstanceKlass (scaled to a count of bytes)
 3595   movl(layout_size, Address(klass, Klass::layout_helper_offset()));
 3596   // test to see if it is malformed in some way
 3597   testl(layout_size, Klass::_lh_instance_slow_path_bit);
 3598   jcc(Assembler::notZero, slow_case_no_pop);
 3599 
 3600   // Allocate the instance:
 3601   //  If TLAB is enabled:
 3602   //    Try to allocate in the TLAB.
 3603   //    If fails, go to the slow path.
 3604   //  Else If inline contiguous allocations are enabled:
 3605   //    Try to allocate in eden.
 3606   //    If fails due to heap end, go to slow path.
 3607   //
 3608   //  If TLAB is enabled OR inline contiguous is enabled:
 3609   //    Initialize the allocation.
 3610   //    Exit.
 3611   //
 3612   //  Go to slow path.
 3613 
 3614   push(klass);
 3615   if (UseTLAB) {
 3616     tlab_allocate(new_obj, layout_size, 0, klass, t2, slow_case);
 3617     if (ZeroTLAB || (!clear_fields)) {
 3618       // the fields have been already cleared
 3619       jmp(initialize_header);
 3620     } else {
 3621       // initialize both the header and fields
 3622       jmp(initialize_object);
 3623     }
 3624   } else {
 3625     jmp(slow_case);
 3626   }
 3627 
 3628   // If UseTLAB is true, the object is created above and there is an initialize need.
 3629   // Otherwise, skip and go to the slow path.
 3630   if (UseTLAB) {
 3631     if (clear_fields) {
 3632       // The object is initialized before the header.  If the object size is
 3633       // zero, go directly to the header initialization.
 3634       bind(initialize_object);
 3635       if (UseCompactObjectHeaders) {
 3636         assert(is_aligned(oopDesc::base_offset_in_bytes(), BytesPerLong), "oop base offset must be 8-byte-aligned");
 3637         decrement(layout_size, oopDesc::base_offset_in_bytes());
 3638       } else {
 3639         decrement(layout_size, sizeof(oopDesc));
 3640       }
 3641       jcc(Assembler::zero, initialize_header);
 3642 
 3643       // Initialize topmost object field, divide size by 8, check if odd and
 3644       // test if zero.
 3645       Register zero = klass;
 3646       xorl(zero, zero);    // use zero reg to clear memory (shorter code)
 3647       shrl(layout_size, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
 3648 
 3649   #ifdef ASSERT
 3650       // make sure instance_size was multiple of 8
 3651       Label L;
 3652       // Ignore partial flag stall after shrl() since it is debug VM
 3653       jcc(Assembler::carryClear, L);
 3654       stop("object size is not multiple of 2 - adjust this code");
 3655       bind(L);
 3656       // must be > 0, no extra check needed here
 3657   #endif
 3658 
 3659       // initialize remaining object fields: instance_size was a multiple of 8
 3660       {
 3661         Label loop;
 3662         bind(loop);
 3663         int header_size_bytes = oopDesc::header_size() * HeapWordSize;
 3664         assert(is_aligned(header_size_bytes, BytesPerLong), "oop header size must be 8-byte-aligned");
 3665         movptr(Address(new_obj, layout_size, Address::times_8, header_size_bytes - 1*oopSize), zero);
 3666         decrement(layout_size);
 3667         jcc(Assembler::notZero, loop);
 3668       }
 3669     } // clear_fields
 3670 
 3671     // initialize object header only.
 3672     bind(initialize_header);
 3673     if (UseCompactObjectHeaders || Arguments::is_valhalla_enabled()) {
 3674       pop(klass);
 3675       Register mark_word = t2;
 3676       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 3677       movptr(Address(new_obj, oopDesc::mark_offset_in_bytes ()), mark_word);
 3678     } else {
 3679      movptr(Address(new_obj, oopDesc::mark_offset_in_bytes()),
 3680             (intptr_t)markWord::prototype().value()); // header
 3681      pop(klass);   // get saved klass back in the register.
 3682     }
 3683     if (!UseCompactObjectHeaders) {
 3684       xorl(rsi, rsi);                 // use zero reg to clear memory (shorter code)
 3685       store_klass_gap(new_obj, rsi);  // zero klass gap for compressed oops
 3686       movptr(t2, klass);         // preserve klass
 3687       store_klass(new_obj, t2, rscratch1);  // src klass reg is potentially compressed
 3688     }
 3689     jmp(done);
 3690   }
 3691 
 3692   bind(slow_case);
 3693   pop(klass);
 3694   bind(slow_case_no_pop);
 3695   jmp(alloc_failed);
 3696 
 3697   bind(done);
 3698 }
 3699 
 3700 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 3701 void MacroAssembler::tlab_allocate(Register obj,
 3702                                    Register var_size_in_bytes,
 3703                                    int con_size_in_bytes,
 3704                                    Register t1,
 3705                                    Register t2,
 3706                                    Label& slow_case) {
 3707   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 3708   bs->tlab_allocate(this, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
 3709 }
 3710 
 3711 RegSet MacroAssembler::call_clobbered_gp_registers() {
 3712   RegSet regs;
 3713   regs += RegSet::of(rax, rcx, rdx);
 3714 #ifndef _WINDOWS
 3715   regs += RegSet::of(rsi, rdi);
 3716 #endif
 3717   regs += RegSet::range(r8, r11);
 3718   if (UseAPX) {
 3719     regs += RegSet::range(r16, as_Register(Register::number_of_registers - 1));

 3883   xorptr(temp, temp);    // use _zero reg to clear memory (shorter code)
 3884   if (UseIncDec) {
 3885     shrptr(index, 3);  // divide by 8/16 and set carry flag if bit 2 was set
 3886   } else {
 3887     shrptr(index, 2);  // use 2 instructions to avoid partial flag stall
 3888     shrptr(index, 1);
 3889   }
 3890 
 3891   // initialize remaining object fields: index is a multiple of 2 now
 3892   {
 3893     Label loop;
 3894     bind(loop);
 3895     movptr(Address(address, index, Address::times_8, offset_in_bytes - 1*BytesPerWord), temp);
 3896     decrement(index);
 3897     jcc(Assembler::notZero, loop);
 3898   }
 3899 
 3900   bind(done);
 3901 }
 3902 
 3903 void MacroAssembler::inline_layout_info(Register holder_klass, Register index, Register layout_info) {
 3904   movptr(layout_info, Address(holder_klass, InstanceKlass::inline_layout_info_array_offset()));
 3905 #ifdef ASSERT
 3906   {
 3907     Label done;
 3908     cmpptr(layout_info, 0);
 3909     jcc(Assembler::notEqual, done);
 3910     stop("inline_layout_info_array is null");
 3911     bind(done);
 3912   }
 3913 #endif
 3914 
 3915   InlineLayoutInfo array[2];
 3916   int size = (char*)&array[1] - (char*)&array[0]; // computing size of array elements
 3917   if (is_power_of_2(size)) {
 3918     shll(index, log2i_exact(size)); // Scale index by power of 2
 3919   } else {
 3920     imull(index, index, size); // Scale the index to be the entry index * array_element_size
 3921   }
 3922   lea(layout_info, Address(layout_info, index, Address::times_1, Array<InlineLayoutInfo>::base_offset_in_bytes()));
 3923 }
 3924 
 3925 // Look up the method for a megamorphic invokeinterface call.
 3926 // The target method is determined by <intf_klass, itable_index>.
 3927 // The receiver klass is in recv_klass.
 3928 // On success, the result will be in method_result, and execution falls through.
 3929 // On failure, execution transfers to the given label.
 3930 void MacroAssembler::lookup_interface_method(Register recv_klass,
 3931                                              Register intf_klass,
 3932                                              RegisterOrConstant itable_index,
 3933                                              Register method_result,
 3934                                              Register scan_temp,
 3935                                              Label& L_no_such_interface,
 3936                                              bool return_method) {
 3937   assert_different_registers(recv_klass, intf_klass, scan_temp);
 3938   assert_different_registers(method_result, intf_klass, scan_temp);
 3939   assert(recv_klass != method_result || !return_method,
 3940          "recv_klass can be destroyed when method isn't needed");
 3941 
 3942   assert(itable_index.is_constant() || itable_index.as_register() == method_result,
 3943          "caller must use same register for non-constant itable index as for method");
 3944 

 4955   } else {
 4956     Label L;
 4957     jccb(negate_condition(cc), L);
 4958     movl(dst, src);
 4959     bind(L);
 4960   }
 4961 }
 4962 
 4963 void MacroAssembler::cmov32(Condition cc, Register dst, Register src) {
 4964   if (VM_Version::supports_cmov()) {
 4965     cmovl(cc, dst, src);
 4966   } else {
 4967     Label L;
 4968     jccb(negate_condition(cc), L);
 4969     movl(dst, src);
 4970     bind(L);
 4971   }
 4972 }
 4973 
 4974 void MacroAssembler::_verify_oop(Register reg, const char* s, const char* file, int line) {
 4975   if (!VerifyOops || VerifyAdapterSharing) {
 4976     // Below address of the code string confuses VerifyAdapterSharing
 4977     // because it may differ between otherwise equivalent adapters.
 4978     return;
 4979   }
 4980 
 4981   BLOCK_COMMENT("verify_oop {");
 4982   push(rscratch1);
 4983   push(rax);                          // save rax
 4984   push(reg);                          // pass register argument
 4985 
 4986   // Pass register number to verify_oop_subroutine
 4987   const char* b = nullptr;
 4988   {
 4989     ResourceMark rm;
 4990     stringStream ss;
 4991     ss.print("verify_oop: %s: %s (%s:%d)", reg->name(), s, file, line);
 4992     b = code_string(ss.as_string());
 4993   }
 4994   AddressLiteral buffer((address) b, external_word_Relocation::spec_for_immediate());
 4995   pushptr(buffer.addr(), rscratch1);
 4996 
 4997   // call indirectly to solve generation ordering problem
 4998   movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address()));
 4999   call(rax);

 5215   // or something else. Since this is a slow path, we can optimize for code density,
 5216   // and just restart the search from the beginning.
 5217   jmpb(L_restart);
 5218 
 5219   // Counter updates:
 5220 
 5221   // Increment polymorphic counter instead of receiver slot.
 5222   bind(L_polymorphic);
 5223   movptr(offset, poly_count_offset);
 5224   jmpb(L_count_update);
 5225 
 5226   // Found a receiver, convert its slot offset to corresponding count offset.
 5227   bind(L_found_recv);
 5228   addptr(offset, receiver_to_count_step);
 5229 
 5230   bind(L_count_update);
 5231   addptr(Address(mdp, offset, Address::times_ptr), DataLayout::counter_increment);
 5232 }
 5233 
 5234 void MacroAssembler::_verify_oop_addr(Address addr, const char* s, const char* file, int line) {
 5235   if (!VerifyOops || VerifyAdapterSharing) {
 5236     // Below address of the code string confuses VerifyAdapterSharing
 5237     // because it may differ between otherwise equivalent adapters.
 5238     return;
 5239   }
 5240 
 5241   push(rscratch1);
 5242   push(rax); // save rax,
 5243   // addr may contain rsp so we will have to adjust it based on the push
 5244   // we just did (and on 64 bit we do two pushes)
 5245   // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which
 5246   // stores rax into addr which is backwards of what was intended.
 5247   if (addr.uses(rsp)) {
 5248     lea(rax, addr);
 5249     pushptr(Address(rax, 2 * BytesPerWord));
 5250   } else {
 5251     pushptr(addr);
 5252   }
 5253 
 5254   // Pass register number to verify_oop_subroutine
 5255   const char* b = nullptr;
 5256   {
 5257     ResourceMark rm;
 5258     stringStream ss;
 5259     ss.print("verify_oop_addr: %s (%s:%d)", s, file, line);

 5613 
 5614 void MacroAssembler::load_mirror(Register mirror, Register method, Register tmp) {
 5615   // get mirror
 5616   const int mirror_offset = in_bytes(Klass::java_mirror_offset());
 5617   load_method_holder(mirror, method);
 5618   movptr(mirror, Address(mirror, mirror_offset));
 5619   resolve_oop_handle(mirror, tmp);
 5620 }
 5621 
 5622 void MacroAssembler::load_method_holder_cld(Register rresult, Register rmethod) {
 5623   load_method_holder(rresult, rmethod);
 5624   movptr(rresult, Address(rresult, InstanceKlass::class_loader_data_offset()));
 5625 }
 5626 
 5627 void MacroAssembler::load_method_holder(Register holder, Register method) {
 5628   movptr(holder, Address(method, Method::const_offset()));                      // ConstMethod*
 5629   movptr(holder, Address(holder, ConstMethod::constants_offset()));             // ConstantPool*
 5630   movptr(holder, Address(holder, ConstantPool::pool_holder_offset()));          // InstanceKlass*
 5631 }
 5632 
 5633 void MacroAssembler::load_metadata(Register dst, Register src) {
 5634   if (UseCompactObjectHeaders) {
 5635     load_narrow_klass_compact(dst, src);
 5636   } else if (UseCompressedClassPointers) {
 5637     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5638   } else {
 5639     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5640   }
 5641 }
 5642 
 5643 void MacroAssembler::load_narrow_klass_compact(Register dst, Register src) {
 5644   assert(UseCompactObjectHeaders, "expect compact object headers");
 5645   movq(dst, Address(src, oopDesc::mark_offset_in_bytes()));
 5646   shrq(dst, markWord::klass_shift);
 5647 }
 5648 
 5649 void MacroAssembler::load_klass(Register dst, Register src, Register tmp) {
 5650   assert_different_registers(src, tmp);
 5651   assert_different_registers(dst, tmp);
 5652 
 5653   if (UseCompactObjectHeaders) {
 5654     load_narrow_klass_compact(dst, src);
 5655     decode_klass_not_null(dst, tmp);
 5656   } else if (UseCompressedClassPointers) {
 5657     movl(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5658     decode_klass_not_null(dst, tmp);
 5659   } else {
 5660     movptr(dst, Address(src, oopDesc::klass_offset_in_bytes()));
 5661   }
 5662 }
 5663 
 5664 void MacroAssembler::load_prototype_header(Register dst, Register src, Register tmp) {
 5665   load_klass(dst, src, tmp);
 5666   movptr(dst, Address(dst, Klass::prototype_header_offset()));
 5667 }
 5668 
 5669 void MacroAssembler::store_klass(Register dst, Register src, Register tmp) {
 5670   assert(!UseCompactObjectHeaders, "not with compact headers");
 5671   assert_different_registers(src, tmp);
 5672   assert_different_registers(dst, tmp);
 5673   if (UseCompressedClassPointers) {
 5674     encode_klass_not_null(src, tmp);
 5675     movl(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5676   } else {
 5677     movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src);
 5678   }
 5679 }
 5680 
 5681 void MacroAssembler::cmp_klass(Register klass, Register obj, Register tmp) {
 5682   if (UseCompactObjectHeaders) {
 5683     assert(tmp != noreg, "need tmp");
 5684     assert_different_registers(klass, obj, tmp);
 5685     load_narrow_klass_compact(tmp, obj);
 5686     cmpl(klass, tmp);
 5687   } else if (UseCompressedClassPointers) {
 5688     cmpl(klass, Address(obj, oopDesc::klass_offset_in_bytes()));

 5714   bool as_raw = (decorators & AS_RAW) != 0;
 5715   if (as_raw) {
 5716     bs->BarrierSetAssembler::load_at(this, decorators, type, dst, src, tmp1);
 5717   } else {
 5718     bs->load_at(this, decorators, type, dst, src, tmp1);
 5719   }
 5720 }
 5721 
 5722 void MacroAssembler::access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 5723                                      Register tmp1, Register tmp2, Register tmp3) {
 5724   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5725   decorators = AccessInternal::decorator_fixup(decorators, type);
 5726   bool as_raw = (decorators & AS_RAW) != 0;
 5727   if (as_raw) {
 5728     bs->BarrierSetAssembler::store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5729   } else {
 5730     bs->store_at(this, decorators, type, dst, val, tmp1, tmp2, tmp3);
 5731   }
 5732 }
 5733 
 5734 void MacroAssembler::flat_field_copy(DecoratorSet decorators, Register src, Register dst,
 5735                                      Register inline_layout_info) {
 5736   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
 5737   bs->flat_field_copy(this, decorators, src, dst, inline_layout_info);
 5738 }
 5739 
 5740 void MacroAssembler::payload_offset(Register inline_klass, Register offset) {
 5741   movptr(offset, Address(inline_klass, InlineKlass::adr_members_offset()));
 5742   movl(offset, Address(offset, InlineKlass::payload_offset_offset()));
 5743 }
 5744 
 5745 void MacroAssembler::payload_addr(Register oop, Register data, Register inline_klass) {
 5746   // ((address) (void*) o) + vk->payload_offset();
 5747   Register offset = (data == oop) ? rscratch1 : data;
 5748   payload_offset(inline_klass, offset);
 5749   if (data == oop) {
 5750     addptr(data, offset);
 5751   } else {
 5752     lea(data, Address(oop, offset));
 5753   }
 5754 }
 5755 
 5756 void MacroAssembler::data_for_value_array_index(Register array, Register array_klass,
 5757                                                 Register index, Register data) {
 5758   assert(index != rcx, "index needs to shift by rcx");
 5759   assert_different_registers(array, array_klass, index);
 5760   assert_different_registers(rcx, array, index);
 5761 
 5762   // array->base() + (index << Klass::layout_helper_log2_element_size(lh));
 5763   movl(rcx, Address(array_klass, Klass::layout_helper_offset()));
 5764 
 5765   // Klass::layout_helper_log2_element_size(lh)
 5766   // (lh >> _lh_log2_element_size_shift) & _lh_log2_element_size_mask;
 5767   shrl(rcx, Klass::_lh_log2_element_size_shift);
 5768   andl(rcx, Klass::_lh_log2_element_size_mask);
 5769   shlptr(index); // index << rcx
 5770 
 5771   lea(data, Address(array, index, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_FLAT_ELEMENT)));
 5772 }
 5773 
 5774 void MacroAssembler::load_heap_oop(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5775   access_load_at(T_OBJECT, IN_HEAP | decorators, dst, src, tmp1);
 5776 }
 5777 
 5778 // Doesn't do verification, generates fixed size code
 5779 void MacroAssembler::load_heap_oop_not_null(Register dst, Address src, Register tmp1, DecoratorSet decorators) {
 5780   access_load_at(T_OBJECT, IN_HEAP | IS_NOT_NULL | decorators, dst, src, tmp1);
 5781 }
 5782 
 5783 void MacroAssembler::store_heap_oop(Address dst, Register val, Register tmp1,
 5784                                     Register tmp2, Register tmp3, DecoratorSet decorators) {
 5785   access_store_at(T_OBJECT, IN_HEAP | decorators, dst, val, tmp1, tmp2, tmp3);
 5786 }
 5787 
 5788 // Used for storing nulls.
 5789 void MacroAssembler::store_heap_oop_null(Address dst) {
 5790   access_store_at(T_OBJECT, IN_HEAP, dst, noreg, noreg, noreg, noreg);
 5791 }
 5792 
 5793 void MacroAssembler::store_klass_gap(Register dst, Register src) {

 6110   assert (oop_recorder() != nullptr, "this assembler needs an OopRecorder");
 6111   int klass_index = oop_recorder()->find_index(k);
 6112   RelocationHolder rspec = metadata_Relocation::spec(klass_index);
 6113   Assembler::cmp_narrow_oop(dst, CompressedKlassPointers::encode(k), rspec);
 6114 }
 6115 
 6116 void MacroAssembler::reinit_heapbase() {
 6117   if (UseCompressedOops) {
 6118     if (Universe::heap() != nullptr) {
 6119       if (CompressedOops::base() == nullptr) {
 6120         MacroAssembler::xorptr(r12_heapbase, r12_heapbase);
 6121       } else {
 6122         mov64(r12_heapbase, (int64_t)CompressedOops::base());
 6123       }
 6124     } else {
 6125       movptr(r12_heapbase, ExternalAddress(CompressedOops::base_addr()));
 6126     }
 6127   }
 6128 }
 6129 
 6130 int MacroAssembler::store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter) {
 6131   assert(InlineTypeReturnedAsFields, "Inline types should never be returned as fields");
 6132   // An inline type might be returned. If fields are in registers we
 6133   // need to allocate an inline type instance and initialize it with
 6134   // the value of the fields.
 6135   Label skip;
 6136   // We only need a new buffered inline type if a new one is not returned
 6137   testptr(rax, 1);
 6138   jcc(Assembler::zero, skip);
 6139   int call_offset = -1;
 6140 
 6141 #ifdef _LP64
 6142   // The following code is similar to allocate_instance but has some slight differences,
 6143   // e.g. object size is always not zero, sometimes it's constant; storing klass ptr after
 6144   // allocating is not necessary if vk != nullptr, etc. allocate_instance is not aware of these.
 6145   Label slow_case;
 6146   // 1. Try to allocate a new buffered inline instance either from TLAB or eden space
 6147   mov(rscratch1, rax); // save rax for slow_case since *_allocate may corrupt it when allocation failed
 6148   if (vk != nullptr) {
 6149     // Called from C1, where the return type is statically known.
 6150     movptr(rbx, (intptr_t)vk->get_InlineKlass());
 6151     jint lh = vk->layout_helper();
 6152     assert(lh != Klass::_lh_neutral_value, "inline class in return type must have been resolved");
 6153     if (UseTLAB && !Klass::layout_helper_needs_slow_path(lh)) {
 6154       tlab_allocate(rax, noreg, lh, r13, r14, slow_case);
 6155     } else {
 6156       jmp(slow_case);
 6157     }
 6158   } else {
 6159     // Call from interpreter. RAX contains ((the InlineKlass* of the return type) | 0x01)
 6160     mov(rbx, rax);
 6161     andptr(rbx, -2);
 6162     if (UseTLAB) {
 6163       movl(r14, Address(rbx, Klass::layout_helper_offset()));
 6164       testl(r14, Klass::_lh_instance_slow_path_bit);
 6165       jcc(Assembler::notZero, slow_case);
 6166       tlab_allocate(rax, r14, 0, r13, r14, slow_case);
 6167     } else {
 6168       jmp(slow_case);
 6169     }
 6170   }
 6171   if (UseTLAB) {
 6172     // 2. Initialize buffered inline instance header
 6173     Register buffer_obj = rax;
 6174     Register klass = rbx;
 6175     if (UseCompactObjectHeaders) {
 6176       Register mark_word = r13;
 6177       movptr(mark_word, Address(klass, Klass::prototype_header_offset()));
 6178       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), mark_word);
 6179     } else {
 6180       movptr(Address(buffer_obj, oopDesc::mark_offset_in_bytes()), (intptr_t)markWord::inline_type_prototype().value());
 6181       xorl(r13, r13);
 6182       store_klass_gap(buffer_obj, r13);
 6183       if (vk == nullptr) {
 6184         // store_klass corrupts rbx(klass), so save it in r13 for later use (interpreter case only).
 6185         mov(r13, klass);
 6186       }
 6187       store_klass(buffer_obj, klass, rscratch1);
 6188       klass = r13;
 6189     }
 6190     // 3. Initialize its fields with an inline class specific handler
 6191     if (vk != nullptr) {
 6192       call(RuntimeAddress(vk->pack_handler())); // no need for call info as this will not safepoint.
 6193     } else {
 6194       movptr(rbx, Address(klass, InlineKlass::adr_members_offset()));
 6195       movptr(rbx, Address(rbx, InlineKlass::pack_handler_offset()));
 6196       call(rbx);
 6197     }
 6198     jmp(skip);
 6199   }
 6200   bind(slow_case);
 6201   // We failed to allocate a new inline type, fall back to a runtime
 6202   // call. Some oop field may be live in some registers but we can't
 6203   // tell. That runtime call will take care of preserving them
 6204   // across a GC if there's one.
 6205   mov(rax, rscratch1);
 6206 #endif
 6207 
 6208   if (from_interpreter) {
 6209     super_call_VM_leaf(StubRoutines::store_inline_type_fields_to_buf());
 6210   } else {
 6211     call(RuntimeAddress(StubRoutines::store_inline_type_fields_to_buf()));
 6212     call_offset = offset();
 6213   }
 6214 
 6215   bind(skip);
 6216   return call_offset;
 6217 }
 6218 
 6219 // Move a value between registers/stack slots and update the reg_state
 6220 bool MacroAssembler::move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]) {
 6221   assert(from->is_valid() && to->is_valid(), "source and destination must be valid");
 6222   if (reg_state[to->value()] == reg_written) {
 6223     return true; // Already written
 6224   }
 6225   if (from != to && bt != T_VOID) {
 6226     if (reg_state[to->value()] == reg_readonly) {
 6227       return false; // Not yet writable
 6228     }
 6229     if (from->is_reg()) {
 6230       if (to->is_reg()) {
 6231         if (from->is_XMMRegister()) {
 6232           if (bt == T_DOUBLE) {
 6233             movdbl(to->as_XMMRegister(), from->as_XMMRegister());
 6234           } else {
 6235             assert(bt == T_FLOAT, "must be float");
 6236             movflt(to->as_XMMRegister(), from->as_XMMRegister());
 6237           }
 6238         } else {
 6239           movq(to->as_Register(), from->as_Register());
 6240         }
 6241       } else {
 6242         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6243         Address to_addr = Address(rsp, st_off);
 6244         if (from->is_XMMRegister()) {
 6245           if (bt == T_DOUBLE) {
 6246             movdbl(to_addr, from->as_XMMRegister());
 6247           } else {
 6248             assert(bt == T_FLOAT, "must be float");
 6249             movflt(to_addr, from->as_XMMRegister());
 6250           }
 6251         } else {
 6252           movq(to_addr, from->as_Register());
 6253         }
 6254       }
 6255     } else {
 6256       Address from_addr = Address(rsp, from->reg2stack() * VMRegImpl::stack_slot_size + wordSize);
 6257       if (to->is_reg()) {
 6258         if (to->is_XMMRegister()) {
 6259           if (bt == T_DOUBLE) {
 6260             movdbl(to->as_XMMRegister(), from_addr);
 6261           } else {
 6262             assert(bt == T_FLOAT, "must be float");
 6263             movflt(to->as_XMMRegister(), from_addr);
 6264           }
 6265         } else {
 6266           movq(to->as_Register(), from_addr);
 6267         }
 6268       } else {
 6269         int st_off = to->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6270         movq(r13, from_addr);
 6271         movq(Address(rsp, st_off), r13);
 6272       }
 6273     }
 6274   }
 6275   // Update register states
 6276   reg_state[from->value()] = reg_writable;
 6277   reg_state[to->value()] = reg_written;
 6278   return true;
 6279 }
 6280 
 6281 // Calculate the extra stack space required for packing or unpacking inline
 6282 // args and adjust the stack pointer (see MacroAssembler::remove_frame).
 6283 int MacroAssembler::extend_stack_for_inline_args(int args_on_stack) {
 6284   int sp_inc = args_on_stack * VMRegImpl::stack_slot_size;
 6285   sp_inc = align_up(sp_inc, StackAlignmentInBytes);
 6286   assert(sp_inc > 0, "sanity");
 6287   // Two additional slots to account for return address
 6288   sp_inc +=  2 * VMRegImpl::stack_slot_size;
 6289 
 6290   push(rbp);
 6291   subptr(rsp, sp_inc);
 6292 #ifdef ASSERT
 6293   movl(Address(rsp, 0), badRegWordVal);
 6294   movl(Address(rsp, VMRegImpl::stack_slot_size), badRegWordVal);
 6295 #endif
 6296   return sp_inc + wordSize; // account for rbp space
 6297 }
 6298 
 6299 // Read all fields from an inline type buffer and store the field values in registers/stack slots.
 6300 bool MacroAssembler::unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
 6301                                           VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
 6302                                           RegState reg_state[]) {
 6303   assert(sig->at(sig_index)._bt == T_VOID, "should be at end delimiter");
 6304   assert(from->is_valid(), "source must be valid");
 6305   bool progress = false;
 6306 #ifdef ASSERT
 6307   const int start_offset = offset();
 6308 #endif
 6309 
 6310   Label L_null, L_notNull;
 6311   // Don't use r14 as tmp because it's used for spilling (see MacroAssembler::spill_reg_for)
 6312   Register tmp1 = r10;
 6313   Register tmp2 = r13;
 6314   Register fromReg = noreg;
 6315   ScalarizedInlineArgsStream stream(sig, sig_index, to, to_count, to_index, true);
 6316   bool done = true;
 6317   bool mark_done = true;
 6318   VMReg toReg;
 6319   BasicType bt;
 6320   // Check if argument requires a null check
 6321   bool null_check = false;
 6322   VMReg nullCheckReg;
 6323   while (stream.next(nullCheckReg, bt)) {
 6324     if (sig->at(stream.sig_index())._offset == -1) {
 6325       null_check = true;
 6326       break;
 6327     }
 6328   }
 6329   stream.reset(sig_index, to_index);
 6330   while (stream.next(toReg, bt)) {
 6331     assert(toReg->is_valid(), "destination must be valid");
 6332     int idx = (int)toReg->value();
 6333     if (reg_state[idx] == reg_readonly) {
 6334       if (idx != from->value()) {
 6335         mark_done = false;
 6336       }
 6337       done = false;
 6338       continue;
 6339     } else if (reg_state[idx] == reg_written) {
 6340       continue;
 6341     }
 6342     assert(reg_state[idx] == reg_writable, "must be writable");
 6343     reg_state[idx] = reg_written;
 6344     progress = true;
 6345 
 6346     if (fromReg == noreg) {
 6347       if (from->is_reg()) {
 6348         fromReg = from->as_Register();
 6349       } else {
 6350         int st_off = from->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6351         movq(tmp1, Address(rsp, st_off));
 6352         fromReg = tmp1;
 6353       }
 6354       if (null_check) {
 6355         // Nullable inline type argument, emit null check
 6356         testptr(fromReg, fromReg);
 6357         jcc(Assembler::zero, L_null);
 6358       }
 6359     }
 6360     int off = sig->at(stream.sig_index())._offset;
 6361     if (off == -1) {
 6362       assert(null_check, "Missing null check at");
 6363       if (toReg->is_stack()) {
 6364         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6365         movq(Address(rsp, st_off), 1);
 6366       } else {
 6367         movq(toReg->as_Register(), 1);
 6368       }
 6369       continue;
 6370     }
 6371     assert(off > 0, "offset in object should be positive");
 6372     Address fromAddr = Address(fromReg, off);
 6373     if (!toReg->is_XMMRegister()) {
 6374       Register dst = toReg->is_stack() ? tmp2 : toReg->as_Register();
 6375       if (is_reference_type(bt)) {
 6376         load_heap_oop(dst, fromAddr);
 6377       } else {
 6378         bool is_signed = (bt != T_CHAR) && (bt != T_BOOLEAN);
 6379         load_sized_value(dst, fromAddr, type2aelembytes(bt), is_signed);
 6380       }
 6381       if (toReg->is_stack()) {
 6382         int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6383         movq(Address(rsp, st_off), dst);
 6384       }
 6385     } else if (bt == T_DOUBLE) {
 6386       movdbl(toReg->as_XMMRegister(), fromAddr);
 6387     } else {
 6388       assert(bt == T_FLOAT, "must be float");
 6389       movflt(toReg->as_XMMRegister(), fromAddr);
 6390     }
 6391   }
 6392   if (progress && null_check) {
 6393     if (done) {
 6394       jmp(L_notNull);
 6395       bind(L_null);
 6396       // Set null marker to zero to signal that the argument is null.
 6397       // Also set all fields to zero since the runtime requires a canonical
 6398       // representation of a flat null.
 6399       stream.reset(sig_index, to_index);
 6400       while (stream.next(toReg, bt)) {
 6401         if (toReg->is_stack()) {
 6402           int st_off = toReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6403           movq(Address(rsp, st_off), 0);
 6404         } else if (toReg->is_XMMRegister()) {
 6405           xorps(toReg->as_XMMRegister(), toReg->as_XMMRegister());
 6406         } else {
 6407           xorl(toReg->as_Register(), toReg->as_Register());
 6408         }
 6409       }
 6410       bind(L_notNull);
 6411     } else {
 6412       bind(L_null);
 6413     }
 6414   }
 6415 
 6416   sig_index = stream.sig_index();
 6417   to_index = stream.regs_index();
 6418 
 6419   if (mark_done && reg_state[from->value()] != reg_written) {
 6420     // This is okay because no one else will write to that slot
 6421     reg_state[from->value()] = reg_writable;
 6422   }
 6423   from_index--;
 6424   assert(progress || (start_offset == offset()), "should not emit code");
 6425   return done;
 6426 }
 6427 
 6428 bool MacroAssembler::pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
 6429                                         VMRegPair* from, int from_count, int& from_index, VMReg to,
 6430                                         RegState reg_state[], Register val_array) {
 6431   assert(sig->at(sig_index)._bt == T_METADATA, "should be at delimiter");
 6432   assert(to->is_valid(), "destination must be valid");
 6433 
 6434   if (reg_state[to->value()] == reg_written) {
 6435     skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6436     return true; // Already written
 6437   }
 6438 
 6439   // TODO 8284443 Isn't it an issue if below code uses r14 as tmp when it contains a spilled value?
 6440   // Be careful with r14 because it's used for spilling (see MacroAssembler::spill_reg_for).
 6441   Register val_obj_tmp = r11;
 6442   Register from_reg_tmp = r14;
 6443   Register tmp1 = r10;
 6444   Register tmp2 = r13;
 6445   Register tmp3 = rbx;
 6446   Register val_obj = to->is_stack() ? val_obj_tmp : to->as_Register();
 6447 
 6448   assert_different_registers(val_obj_tmp, from_reg_tmp, tmp1, tmp2, tmp3, val_array);
 6449 
 6450   if (reg_state[to->value()] == reg_readonly) {
 6451     if (!is_reg_in_unpacked_fields(sig, sig_index, to, from, from_count, from_index)) {
 6452       skip_unpacked_fields(sig, sig_index, from, from_count, from_index);
 6453       return false; // Not yet writable
 6454     }
 6455     val_obj = val_obj_tmp;
 6456   }
 6457 
 6458   int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + vtarg_index * type2aelembytes(T_OBJECT);
 6459   load_heap_oop(val_obj, Address(val_array, index));
 6460 
 6461   ScalarizedInlineArgsStream stream(sig, sig_index, from, from_count, from_index);
 6462   VMReg fromReg;
 6463   BasicType bt;
 6464   Label L_null;
 6465   while (stream.next(fromReg, bt)) {
 6466     assert(fromReg->is_valid(), "source must be valid");
 6467     reg_state[fromReg->value()] = reg_writable;
 6468 
 6469     int off = sig->at(stream.sig_index())._offset;
 6470     if (off == -1) {
 6471       // Nullable inline type argument, emit null check
 6472       Label L_notNull;
 6473       if (fromReg->is_stack()) {
 6474         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6475         testb(Address(rsp, ld_off), 1);
 6476       } else {
 6477         testb(fromReg->as_Register(), 1);
 6478       }
 6479       jcc(Assembler::notZero, L_notNull);
 6480       movptr(val_obj, 0);
 6481       jmp(L_null);
 6482       bind(L_notNull);
 6483       continue;
 6484     }
 6485 
 6486     assert(off > 0, "offset in object should be positive");
 6487     size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
 6488 
 6489     // Pack the scalarized field into the value object.
 6490     Address dst(val_obj, off);
 6491     if (!fromReg->is_XMMRegister()) {
 6492       Register src;
 6493       if (fromReg->is_stack()) {
 6494         src = from_reg_tmp;
 6495         int ld_off = fromReg->reg2stack() * VMRegImpl::stack_slot_size + wordSize;
 6496         load_sized_value(src, Address(rsp, ld_off), size_in_bytes, /* is_signed */ false);
 6497       } else {
 6498         src = fromReg->as_Register();
 6499       }
 6500       assert_different_registers(dst.base(), src, tmp1, tmp2, tmp3, val_array);
 6501       if (is_reference_type(bt)) {
 6502         // store_heap_oop transitively calls oop_store_at which corrupts to.base(). We need to keep val_obj valid.
 6503         mov(tmp3, val_obj);
 6504         Address dst_with_tmp3(tmp3, off);
 6505         store_heap_oop(dst_with_tmp3, src, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
 6506       } else {
 6507         store_sized_value(dst, src, size_in_bytes);
 6508       }
 6509     } else if (bt == T_DOUBLE) {
 6510       movdbl(dst, fromReg->as_XMMRegister());
 6511     } else {
 6512       assert(bt == T_FLOAT, "must be float");
 6513       movflt(dst, fromReg->as_XMMRegister());
 6514     }
 6515   }
 6516   bind(L_null);
 6517   sig_index = stream.sig_index();
 6518   from_index = stream.regs_index();
 6519 
 6520   assert(reg_state[to->value()] == reg_writable, "must have already been read");
 6521   bool success = move_helper(val_obj->as_VMReg(), to, T_OBJECT, reg_state);
 6522   assert(success, "to register must be writeable");
 6523   return true;
 6524 }
 6525 
 6526 VMReg MacroAssembler::spill_reg_for(VMReg reg) {
 6527   return reg->is_XMMRegister() ? xmm8->as_VMReg() : r14->as_VMReg();
 6528 }
 6529 
 6530 void MacroAssembler::remove_frame(int initial_framesize, bool needs_stack_repair) {
 6531   assert((initial_framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
 6532   if (needs_stack_repair) {
 6533     // The method has a scalarized entry point (where fields of value object arguments
 6534     // are passed through registers and stack), and a non-scalarized entry point (where
 6535     // value object arguments are given as oops). The non-scalarized entry point will
 6536     // first load each field of value object arguments and store them in registers and on
 6537     // the stack in a way compatible with the scalarized entry point. To do so, some extra
 6538     // stack space might be reserved (if argument registers are not enough). On leaving the
 6539     // method, this space must be freed.
 6540     //
 6541     // In case we used the non-scalarized entry point the stack looks like this:
 6542     //
 6543     // | Arguments from caller     |
 6544     // |---------------------------|  <-- caller's SP
 6545     // | Return address #1         |
 6546     // | Saved RBP #1              |
 6547     // |---------------------------|
 6548     // | Extension space for       |
 6549     // |   inline arg (un)packing  |
 6550     // |---------------------------|  <-- start of this method's frame
 6551     // | Return address #2         |
 6552     // | Saved RBP #2              |
 6553     // |---------------------------|  <-- RBP (with -XX:+PreserveFramePointer)
 6554     // | sp_inc                    |
 6555     // | method locals             |
 6556     // |---------------------------|  <-- SP
 6557     //
 6558     // Space for the return pc and saved rbp is reserved twice. But only the #1 copies
 6559     // contain the real values of return pc and saved rbp. The #2 copies are not reliable
 6560     // and should not be used. They are mostly needed to add space between the  extension
 6561     // space and the locals, as there would be between the real arguments and the locals
 6562     // if we don't need to do unpacking (from the scalarized entry point).
 6563     //
 6564     // When leaving, one must load RBP #1 into RBP, and use the copy #1 of the return address,
 6565     // while keeping in mind that from the scalarized entry point, there will be only one
 6566     // copy. Indeed, in the case we used the scalarized calling convention, the stack looks like this:
 6567     //
 6568     // | Arguments from caller     |
 6569     // |---------------------------|  <-- caller's SP
 6570     // | Return address            |
 6571     // | Saved RBP                 |
 6572     // |---------------------------|  <-- FP (with -XX:+PreserveFramePointer)
 6573     // | sp_inc                    |
 6574     // | method locals             |
 6575     // |---------------------------|  <-- SP
 6576     //
 6577     // The sp_inc stack slot holds the total size of the frame, including the extension
 6578     // space and copies #2 of the return address and the saved RBP (but never the copies
 6579     // #1 of the return address and saved RBP). That is how to find the copies #1 of the
 6580     // return address and saved rbp. This size is expressed in bytes. Be careful when using
 6581     // it from C++ in pointer arithmetic you might need to divide it by wordSize.
 6582 
 6583     // The stack increment resides just below the saved rbp
 6584     addq(rsp, Address(rsp, initial_framesize - wordSize));
 6585     pop(rbp);
 6586   } else {
 6587     if (initial_framesize > 0) {
 6588       addq(rsp, initial_framesize);
 6589     }
 6590     pop(rbp);
 6591   }
 6592 }
 6593 
 6594 #if COMPILER2_OR_JVMCI
 6595 
 6596 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM/ZMM registers
 6597 void MacroAssembler::xmm_clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, KRegister mask) {
 6598   // cnt - number of qwords (8-byte words).
 6599   // base - start address, qword aligned.
 6600   Label L_zero_64_bytes, L_loop, L_sloop, L_tail, L_end;
 6601   bool use64byteVector = (MaxVectorSize == 64) && (VM_Version::avx3_threshold() == 0);
 6602   if (use64byteVector) {
 6603     evpbroadcastq(xtmp, val, AVX_512bit);
 6604   } else if (MaxVectorSize >= 32) {
 6605     movdq(xtmp, val);
 6606     punpcklqdq(xtmp, xtmp);
 6607     vinserti128_high(xtmp, xtmp);
 6608   } else {
 6609     movdq(xtmp, val);
 6610     punpcklqdq(xtmp, xtmp);
 6611   }
 6612   jmp(L_zero_64_bytes);
 6613 
 6614   BIND(L_loop);
 6615   if (MaxVectorSize >= 32) {
 6616     fill64(base, 0, xtmp, use64byteVector);
 6617   } else {
 6618     movdqu(Address(base,  0), xtmp);
 6619     movdqu(Address(base, 16), xtmp);
 6620     movdqu(Address(base, 32), xtmp);
 6621     movdqu(Address(base, 48), xtmp);
 6622   }
 6623   addptr(base, 64);
 6624 
 6625   BIND(L_zero_64_bytes);
 6626   subptr(cnt, 8);
 6627   jccb(Assembler::greaterEqual, L_loop);
 6628 
 6629   // Copy trailing 64 bytes
 6630   if (use64byteVector) {
 6631     addptr(cnt, 8);
 6632     jccb(Assembler::equal, L_end);
 6633     fill64_masked(3, base, 0, xtmp, mask, cnt, val, true);
 6634     jmp(L_end);
 6635   } else {
 6636     addptr(cnt, 4);
 6637     jccb(Assembler::less, L_tail);
 6638     if (MaxVectorSize >= 32) {
 6639       vmovdqu(Address(base, 0), xtmp);
 6640     } else {
 6641       movdqu(Address(base,  0), xtmp);
 6642       movdqu(Address(base, 16), xtmp);
 6643     }
 6644   }
 6645   addptr(base, 32);
 6646   subptr(cnt, 4);
 6647 
 6648   BIND(L_tail);
 6649   addptr(cnt, 4);
 6650   jccb(Assembler::lessEqual, L_end);
 6651   if (UseAVX > 2 && MaxVectorSize >= 32 && VM_Version::supports_avx512vl()) {
 6652     fill32_masked(3, base, 0, xtmp, mask, cnt, val);
 6653   } else {
 6654     decrement(cnt);
 6655 
 6656     BIND(L_sloop);
 6657     movq(Address(base, 0), xtmp);
 6658     addptr(base, 8);
 6659     decrement(cnt);
 6660     jccb(Assembler::greaterEqual, L_sloop);
 6661   }
 6662   BIND(L_end);
 6663 }
 6664 
 6665 // Clearing constant sized memory using YMM/ZMM registers.
 6666 void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask) {
 6667   assert(UseAVX > 2 && VM_Version::supports_avx512vl(), "");
 6668   bool use64byteVector = (MaxVectorSize > 32) && (VM_Version::avx3_threshold() == 0);
 6669 
 6670   int vector64_count = (cnt & (~0x7)) >> 3;
 6671   cnt = cnt & 0x7;
 6672   const int fill64_per_loop = 4;

 6734         break;
 6735       case 7:
 6736         if (use64byteVector) {
 6737           movl(rtmp, 0x7F);
 6738           kmovwl(mask, rtmp);
 6739           evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
 6740         } else {
 6741           evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
 6742           movl(rtmp, 0x7);
 6743           kmovwl(mask, rtmp);
 6744           evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
 6745         }
 6746         break;
 6747       default:
 6748         fatal("Unexpected length : %d\n",cnt);
 6749         break;
 6750     }
 6751   }
 6752 }
 6753 
 6754 void MacroAssembler::clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp,
 6755                                bool is_large, bool word_copy_only, KRegister mask) {
 6756   // cnt      - number of qwords (8-byte words).
 6757   // base     - start address, qword aligned.
 6758   // is_large - if optimizers know cnt is larger than InitArrayShortSize
 6759   assert(base==rdi, "base register must be edi for rep stos");
 6760   assert(val==rax,   "val register must be eax for rep stos");
 6761   assert(cnt==rcx,   "cnt register must be ecx for rep stos");
 6762   assert(InitArrayShortSize % BytesPerLong == 0,
 6763     "InitArrayShortSize should be the multiple of BytesPerLong");
 6764 
 6765   Label DONE;



 6766 
 6767   if (!is_large) {
 6768     Label LOOP, LONG;
 6769     cmpptr(cnt, InitArrayShortSize/BytesPerLong);
 6770     jccb(Assembler::greater, LONG);
 6771 
 6772     decrement(cnt);
 6773     jccb(Assembler::negative, DONE); // Zero length
 6774 
 6775     // Use individual pointer-sized stores for small counts:
 6776     BIND(LOOP);
 6777     movptr(Address(base, cnt, Address::times_ptr), val);
 6778     decrement(cnt);
 6779     jccb(Assembler::greaterEqual, LOOP);
 6780     jmpb(DONE);
 6781 
 6782     BIND(LONG);
 6783   }
 6784 
 6785   // Use longer rep-prefixed ops for non-small counts:
 6786   if (UseFastStosb && !word_copy_only) {
 6787     shlptr(cnt, 3); // convert to number of bytes
 6788     rep_stosb();
 6789   } else if (UseXMMForObjInit) {
 6790     xmm_clear_mem(base, cnt, val, xtmp, mask);
 6791   } else {
 6792     rep_stos();
 6793   }
 6794 
 6795   BIND(DONE);
 6796 }
 6797 
 6798 #endif //COMPILER2_OR_JVMCI
 6799 
 6800 
 6801 void MacroAssembler::generate_fill(BasicType t, bool aligned,
 6802                                    Register to, Register value, Register count,
 6803                                    Register rtmp, XMMRegister xtmp) {
 6804   ShortBranchVerifier sbv(this);
 6805   assert_different_registers(to, value, count, rtmp);
 6806   Label L_exit;
 6807   Label L_fill_2_bytes, L_fill_4_bytes;
 6808 
 6809 #if defined(COMPILER2)
 6810   if(MaxVectorSize >=32 &&

10690 
10691   // Load top.
10692   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10693 
10694   // Check if the lock-stack is full.
10695   cmpl(top, LockStack::end_offset());
10696   jcc(Assembler::greaterEqual, slow);
10697 
10698   // Check for recursion.
10699   cmpptr(obj, Address(thread, top, Address::times_1, -oopSize));
10700   jcc(Assembler::equal, push);
10701 
10702   // Check header for monitor (0b10).
10703   testptr(reg_rax, markWord::monitor_value);
10704   jcc(Assembler::notZero, slow);
10705 
10706   // Try to lock. Transition lock bits 0b01 => 0b00
10707   movptr(tmp, reg_rax);
10708   andptr(tmp, ~(int32_t)markWord::unlocked_value);
10709   orptr(reg_rax, markWord::unlocked_value);
10710   // Mask inline_type bit such that we go to the slow path if object is an inline type
10711   andptr(reg_rax, ~((int) markWord::inline_type_bit_in_place));
10712 
10713   lock(); cmpxchgptr(tmp, Address(obj, oopDesc::mark_offset_in_bytes()));
10714   jcc(Assembler::notEqual, slow);
10715 
10716   // Restore top, CAS clobbers register.
10717   movl(top, Address(thread, JavaThread::lock_stack_top_offset()));
10718 
10719   bind(push);
10720   // After successful lock, push object on lock-stack.
10721   movptr(Address(thread, top), obj);
10722   incrementl(top, oopSize);
10723   movl(Address(thread, JavaThread::lock_stack_top_offset()), top);
10724 }
10725 
10726 // Implements fast-unlocking.
10727 //
10728 // obj: the object to be unlocked
10729 // reg_rax: rax
10730 // thread: the thread
10731 // tmp: a temporary register
10732 void MacroAssembler::fast_unlock(Register obj, Register reg_rax, Register tmp, Label& slow) {
< prev index next >