< prev index next >

src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp

Print this page

  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  28 
  29 #include "asm/assembler.inline.hpp"
  30 #include "code/aotCodeCache.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "oops/compressedKlass.hpp"
  35 #include "runtime/vm_version.hpp"

  36 #include "utilities/powerOfTwo.hpp"




  37 
  38 class OopMap;
  39 
  40 // MacroAssembler extends Assembler by frequently used macros.
  41 //
  42 // Instructions for which a 'better' code sequence exists depending
  43 // on arguments should also go in here.
  44 
  45 class MacroAssembler: public Assembler {
  46   friend class LIR_Assembler;
  47 
  48  public:
  49   using Assembler::mov;
  50   using Assembler::movi;
  51 
  52  protected:
  53 
  54   // Support for VM calls
  55   //
  56   // This is the base routine called by the different versions of call_VM_leaf. The interpreter

 167 
 168   void bind(Label& L) {
 169     Assembler::bind(L);
 170     code()->clear_last_insn();
 171     code()->set_last_label(pc());
 172   }
 173 
 174   void membar(Membar_mask_bits order_constraint);
 175 
 176   using Assembler::ldr;
 177   using Assembler::str;
 178   using Assembler::ldrw;
 179   using Assembler::strw;
 180 
 181   void ldr(Register Rx, const Address &adr);
 182   void ldrw(Register Rw, const Address &adr);
 183   void str(Register Rx, const Address &adr);
 184   void strw(Register Rx, const Address &adr);
 185 
 186   // Frame creation and destruction shared between JITs.
 187   void build_frame(int framesize);

 188   void remove_frame(int framesize);
 189 
 190   virtual void _call_Unimplemented(address call_site) {
 191     mov(rscratch2, call_site);
 192   }
 193 
 194 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 195 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 196 // https://reviews.llvm.org/D3311
 197 
 198 #ifdef _WIN64
 199 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 200 #else
 201 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 202 #endif
 203 
 204   // aliases defined in AARCH64 spec
 205 
 206   template<class T>
 207   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }

 668     mrs(0b011, 0b1110, 0b0000, 0b110, reg);
 669   }
 670 
 671   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 672   int corrected_idivl(Register result, Register ra, Register rb,
 673                       bool want_remainder, Register tmp = rscratch1);
 674   int corrected_idivq(Register result, Register ra, Register rb,
 675                       bool want_remainder, Register tmp = rscratch1);
 676 
 677   // Support for null-checks
 678   //
 679   // Generates code that causes a null OS exception if the content of reg is null.
 680   // If the accessed location is M[reg + offset] and the offset is known, provide the
 681   // offset. No explicit code generation is needed if the offset is within a certain
 682   // range (0 <= offset <= page_size).
 683 
 684   virtual void null_check(Register reg, int offset = -1);
 685   static bool needs_explicit_null_check(intptr_t offset);
 686   static bool uses_implicit_null_check(void* address);
 687 






















 688   static address target_addr_for_insn(address insn_addr);
 689 
 690   // Required platform-specific helpers for Label::patch_instructions.
 691   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 692   static int pd_patch_instruction_size(address branch, address target);
 693   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 694     pd_patch_instruction_size(branch, target);
 695   }
 696   static address pd_call_destination(address branch) {
 697     return target_addr_for_insn(branch);
 698   }
 699 #ifndef PRODUCT
 700   static void pd_print_patched_instruction(address branch);
 701 #endif
 702 
 703   static int patch_oop(address insn_addr, address o);
 704   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 705 
 706   // Return whether code is emitted to a scratch blob.
 707   virtual bool in_scratch_emit_size() {

 896 
 897   void reset_last_Java_frame(Register thread);
 898 
 899   // thread in the default location (rthread)
 900   void reset_last_Java_frame(bool clear_fp);
 901 
 902   // Stores
 903   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 904   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 905 
 906   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 907   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 908 
 909   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 910   void c2bool(Register x);
 911 
 912   void load_method_holder_cld(Register rresult, Register rmethod);
 913   void load_method_holder(Register holder, Register method);
 914 
 915   // oop manipulations


 916   void load_narrow_klass_compact(Register dst, Register src);
 917   void load_klass(Register dst, Register src);
 918   void store_klass(Register dst, Register src);
 919   void cmp_klass(Register obj, Register klass, Register tmp);
 920   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 921 
 922   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 923   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 924   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 925 
 926   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 927                       Register tmp1, Register tmp2);
 928 
 929   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 930                        Register tmp1, Register tmp2, Register tmp3);
 931 









 932   void load_heap_oop(Register dst, Address src, Register tmp1,
 933                      Register tmp2, DecoratorSet decorators = 0);
 934 
 935   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 936                               Register tmp2, DecoratorSet decorators = 0);
 937   void store_heap_oop(Address dst, Register val, Register tmp1,
 938                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 939 
 940   // currently unimplemented
 941   // Used for storing null. All other oop constants should be
 942   // stored using routines that take a jobject.
 943   void store_heap_oop_null(Address dst);
 944 


 945   void store_klass_gap(Register dst, Register src);
 946 
 947   // This dummy is to prevent a call to store_heap_oop from
 948   // converting a zero (like null) into a Register by giving
 949   // the compiler two choices it can't resolve
 950 
 951   void store_heap_oop(Address dst, void* dummy);
 952 
 953   void encode_heap_oop(Register d, Register s);
 954   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 955   void decode_heap_oop(Register d, Register s);
 956   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 957   void encode_heap_oop_not_null(Register r);
 958   void decode_heap_oop_not_null(Register r);
 959   void encode_heap_oop_not_null(Register dst, Register src);
 960   void decode_heap_oop_not_null(Register dst, Register src);
 961 
 962   void set_narrow_oop(Register dst, jobject obj);
 963 
 964   void decode_klass_not_null_for_aot(Register dst, Register src);

 974   void reinit_heapbase();
 975 
 976   DEBUG_ONLY(void verify_heapbase(const char* msg);)
 977 
 978   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
 979                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 980   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
 981                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
 982 
 983   void push_cont_fastpath(Register java_thread = rthread);
 984   void pop_cont_fastpath(Register java_thread = rthread);
 985 
 986   // Round up to a power of two
 987   void round_to(Register reg, int modulus);
 988 
 989   // java.lang.Math::round intrinsics
 990   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
 991   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
 992 
 993   // allocation









 994   void tlab_allocate(
 995     Register obj,                      // result: pointer to object after successful allocation
 996     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
 997     int      con_size_in_bytes,        // object size in bytes if   known at compile time
 998     Register t1,                       // temp register
 999     Register t2,                       // temp register
1000     Label&   slow_case                 // continuation point if fast allocation fails
1001   );
1002   void verify_tlab();
1003 


1004   // interface method calling
1005   void lookup_interface_method(Register recv_klass,
1006                                Register intf_klass,
1007                                RegisterOrConstant itable_index,
1008                                Register method_result,
1009                                Register scan_temp,
1010                                Label& no_such_interface,
1011                    bool return_method = true);
1012 
1013   void lookup_interface_method_stub(Register recv_klass,
1014                                     Register holder_klass,
1015                                     Register resolved_klass,
1016                                     Register method_result,
1017                                     Register temp_reg,
1018                                     Register temp_reg2,
1019                                     int itable_index,
1020                                     Label& L_no_such_interface);
1021 
1022   // virtual method calling
1023   // n.b. x86 allows RegisterOrConstant for vtable_index

1441   }                                                                     \
1442                                                                         \
1443   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1444     Assembler::INSN(Rd, Rn, Rm);                                        \
1445   }                                                                     \
1446                                                                         \
1447   void INSN(Register Rd, Register Rn, Register Rm,                      \
1448            ext::operation option, int amount = 0) {                     \
1449     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1450   }
1451 
1452   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1453 
1454   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1455   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1456   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1457   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1458 
1459   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1460 


















1461   void tableswitch(Register index, jint lowbound, jint highbound,
1462                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1463     adr(rscratch1, jumptable);
1464     subsw(rscratch2, index, lowbound);
1465     subsw(zr, rscratch2, highbound - lowbound);
1466     br(Assembler::HS, jumptable_end);
1467     add(rscratch1, rscratch1, rscratch2,
1468         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1469     br(rscratch1);
1470   }
1471 
1472   // Form an address from base + offset in Rd.  Rd may or may not
1473   // actually be used: you must use the Address that is returned.  It
1474   // is up to you to ensure that the shift provided matches the size
1475   // of your data.
1476   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1477 
1478   // Return true iff an address is within the 48-bit AArch64 address
1479   // space.
1480   bool is_valid_AArch64_address(address a) {

1515 #define ARRAYS_HASHCODE_REGISTERS \
1516   do {                      \
1517     assert(result == r0  && \
1518            ary    == r1  && \
1519            cnt    == r2  && \
1520            vdata0 == v3  && \
1521            vdata1 == v2  && \
1522            vdata2 == v1  && \
1523            vdata3 == v0  && \
1524            vmul0  == v4  && \
1525            vmul1  == v5  && \
1526            vmul2  == v6  && \
1527            vmul3  == v7  && \
1528            vpow   == v12 && \
1529            vpowm  == v13, "registers must match aarch64.ad"); \
1530   } while (0)
1531 
1532   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1533 
1534   void fill_words(Register base, Register cnt, Register value);


1535   address zero_words(Register base, uint64_t cnt);
1536   address zero_words(Register ptr, Register cnt);
1537   void zero_dcache_blocks(Register base, Register cnt);
1538 
1539   static const int zero_words_block_size;
1540 
1541   address byte_array_inflate(Register src, Register dst, Register len,
1542                              FloatRegister vtmp1, FloatRegister vtmp2,
1543                              FloatRegister vtmp3, Register tmp4);
1544 
1545   void char_array_compress(Register src, Register dst, Register len,
1546                            Register res,
1547                            FloatRegister vtmp0, FloatRegister vtmp1,
1548                            FloatRegister vtmp2, FloatRegister vtmp3,
1549                            FloatRegister vtmp4, FloatRegister vtmp5);
1550 
1551   void encode_iso_array(Register src, Register dst,
1552                         Register len, Register res, bool ascii,
1553                         FloatRegister vtmp0, FloatRegister vtmp1,
1554                         FloatRegister vtmp2, FloatRegister vtmp3,

  16  * You should have received a copy of the GNU General Public License version
  17  * 2 along with this work; if not, write to the Free Software Foundation,
  18  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  21  * or visit www.oracle.com if you need additional information or have any
  22  * questions.
  23  *
  24  */
  25 
  26 #ifndef CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  27 #define CPU_AARCH64_MACROASSEMBLER_AARCH64_HPP
  28 
  29 #include "asm/assembler.inline.hpp"
  30 #include "code/aotCodeCache.hpp"
  31 #include "code/vmreg.hpp"
  32 #include "metaprogramming/enableIf.hpp"
  33 #include "oops/compressedOops.hpp"
  34 #include "oops/compressedKlass.hpp"
  35 #include "runtime/vm_version.hpp"
  36 #include "utilities/macros.hpp"
  37 #include "utilities/powerOfTwo.hpp"
  38 #include "runtime/signature.hpp"
  39 
  40 
  41 class ciInlineKlass;
  42 
  43 class OopMap;
  44 
  45 // MacroAssembler extends Assembler by frequently used macros.
  46 //
  47 // Instructions for which a 'better' code sequence exists depending
  48 // on arguments should also go in here.
  49 
  50 class MacroAssembler: public Assembler {
  51   friend class LIR_Assembler;
  52 
  53  public:
  54   using Assembler::mov;
  55   using Assembler::movi;
  56 
  57  protected:
  58 
  59   // Support for VM calls
  60   //
  61   // This is the base routine called by the different versions of call_VM_leaf. The interpreter

 172 
 173   void bind(Label& L) {
 174     Assembler::bind(L);
 175     code()->clear_last_insn();
 176     code()->set_last_label(pc());
 177   }
 178 
 179   void membar(Membar_mask_bits order_constraint);
 180 
 181   using Assembler::ldr;
 182   using Assembler::str;
 183   using Assembler::ldrw;
 184   using Assembler::strw;
 185 
 186   void ldr(Register Rx, const Address &adr);
 187   void ldrw(Register Rw, const Address &adr);
 188   void str(Register Rx, const Address &adr);
 189   void strw(Register Rx, const Address &adr);
 190 
 191   // Frame creation and destruction shared between JITs.
 192   DEBUG_ONLY(void build_frame(int framesize);)
 193   void build_frame(int framesize DEBUG_ONLY(COMMA bool zap_rfp_lr_spills));
 194   void remove_frame(int framesize);
 195 
 196   virtual void _call_Unimplemented(address call_site) {
 197     mov(rscratch2, call_site);
 198   }
 199 
 200 // Microsoft's MSVC team thinks that the __FUNCSIG__ is approximately (sympathy for calling conventions) equivalent to __PRETTY_FUNCTION__
 201 // Also, from Clang patch: "It is very similar to GCC's PRETTY_FUNCTION, except it prints the calling convention."
 202 // https://reviews.llvm.org/D3311
 203 
 204 #ifdef _WIN64
 205 #define call_Unimplemented() _call_Unimplemented((address)__FUNCSIG__)
 206 #else
 207 #define call_Unimplemented() _call_Unimplemented((address)__PRETTY_FUNCTION__)
 208 #endif
 209 
 210   // aliases defined in AARCH64 spec
 211 
 212   template<class T>
 213   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }

 674     mrs(0b011, 0b1110, 0b0000, 0b110, reg);
 675   }
 676 
 677   // idiv variant which deals with MINLONG as dividend and -1 as divisor
 678   int corrected_idivl(Register result, Register ra, Register rb,
 679                       bool want_remainder, Register tmp = rscratch1);
 680   int corrected_idivq(Register result, Register ra, Register rb,
 681                       bool want_remainder, Register tmp = rscratch1);
 682 
 683   // Support for null-checks
 684   //
 685   // Generates code that causes a null OS exception if the content of reg is null.
 686   // If the accessed location is M[reg + offset] and the offset is known, provide the
 687   // offset. No explicit code generation is needed if the offset is within a certain
 688   // range (0 <= offset <= page_size).
 689 
 690   virtual void null_check(Register reg, int offset = -1);
 691   static bool needs_explicit_null_check(intptr_t offset);
 692   static bool uses_implicit_null_check(void* address);
 693 
 694   // markWord tests, kills markWord reg
 695   void test_markword_is_inline_type(Register markword, Label& is_inline_type);
 696 
 697   // inlineKlass queries, kills temp_reg
 698   void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
 699 
 700   void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
 701   void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
 702   void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
 703   void test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker);
 704 
 705   // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
 706   void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
 707   void test_flat_array_oop(Register klass, Register temp_reg, Label& is_flat_array);
 708   void test_non_flat_array_oop(Register oop, Register temp_reg, Label&is_non_flat_array);
 709   void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
 710   void test_non_null_free_array_oop(Register oop, Register temp_reg, Label&is_non_null_free_array);
 711 
 712   // Check array klass layout helper for flat or null-free arrays...
 713   void test_flat_array_layout(Register lh, Label& is_flat_array);
 714   void test_non_flat_array_layout(Register lh, Label& is_non_flat_array);
 715 
 716   static address target_addr_for_insn(address insn_addr);
 717 
 718   // Required platform-specific helpers for Label::patch_instructions.
 719   // They _shadow_ the declarations in AbstractAssembler, which are undefined.
 720   static int pd_patch_instruction_size(address branch, address target);
 721   static void pd_patch_instruction(address branch, address target, const char* file = nullptr, int line = 0) {
 722     pd_patch_instruction_size(branch, target);
 723   }
 724   static address pd_call_destination(address branch) {
 725     return target_addr_for_insn(branch);
 726   }
 727 #ifndef PRODUCT
 728   static void pd_print_patched_instruction(address branch);
 729 #endif
 730 
 731   static int patch_oop(address insn_addr, address o);
 732   static int patch_narrow_klass(address insn_addr, narrowKlass n);
 733 
 734   // Return whether code is emitted to a scratch blob.
 735   virtual bool in_scratch_emit_size() {

 924 
 925   void reset_last_Java_frame(Register thread);
 926 
 927   // thread in the default location (rthread)
 928   void reset_last_Java_frame(bool clear_fp);
 929 
 930   // Stores
 931   void store_check(Register obj);                // store check for obj - register is destroyed afterwards
 932   void store_check(Register obj, Address dst);   // same as above, dst is exact store location (reg. is destroyed)
 933 
 934   void resolve_jobject(Register value, Register tmp1, Register tmp2);
 935   void resolve_global_jobject(Register value, Register tmp1, Register tmp2);
 936 
 937   // C 'boolean' to Java boolean: x == 0 ? 0 : 1
 938   void c2bool(Register x);
 939 
 940   void load_method_holder_cld(Register rresult, Register rmethod);
 941   void load_method_holder(Register holder, Register method);
 942 
 943   // oop manipulations
 944   void load_metadata(Register dst, Register src);
 945 
 946   void load_narrow_klass_compact(Register dst, Register src);
 947   void load_klass(Register dst, Register src);
 948   void store_klass(Register dst, Register src);
 949   void cmp_klass(Register obj, Register klass, Register tmp);
 950   void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
 951 
 952   void resolve_weak_handle(Register result, Register tmp1, Register tmp2);
 953   void resolve_oop_handle(Register result, Register tmp1, Register tmp2);
 954   void load_mirror(Register dst, Register method, Register tmp1, Register tmp2);
 955 
 956   void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
 957                       Register tmp1, Register tmp2);
 958 
 959   void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
 960                        Register tmp1, Register tmp2, Register tmp3);
 961 
 962   void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
 963 
 964   // inline type data payload offsets...
 965   void payload_offset(Register inline_klass, Register offset);
 966   void payload_address(Register oop, Register data, Register inline_klass);
 967   // get data payload ptr a flat value array at index, kills rcx and index
 968   void data_for_value_array_index(Register array, Register array_klass,
 969                                   Register index, Register data);
 970 
 971   void load_heap_oop(Register dst, Address src, Register tmp1,
 972                      Register tmp2, DecoratorSet decorators = 0);
 973 
 974   void load_heap_oop_not_null(Register dst, Address src, Register tmp1,
 975                               Register tmp2, DecoratorSet decorators = 0);
 976   void store_heap_oop(Address dst, Register val, Register tmp1,
 977                       Register tmp2, Register tmp3, DecoratorSet decorators = 0);
 978 
 979   // currently unimplemented
 980   // Used for storing null. All other oop constants should be
 981   // stored using routines that take a jobject.
 982   void store_heap_oop_null(Address dst);
 983 
 984   void load_prototype_header(Register dst, Register src);
 985 
 986   void store_klass_gap(Register dst, Register src);
 987 
 988   // This dummy is to prevent a call to store_heap_oop from
 989   // converting a zero (like null) into a Register by giving
 990   // the compiler two choices it can't resolve
 991 
 992   void store_heap_oop(Address dst, void* dummy);
 993 
 994   void encode_heap_oop(Register d, Register s);
 995   void encode_heap_oop(Register r) { encode_heap_oop(r, r); }
 996   void decode_heap_oop(Register d, Register s);
 997   void decode_heap_oop(Register r) { decode_heap_oop(r, r); }
 998   void encode_heap_oop_not_null(Register r);
 999   void decode_heap_oop_not_null(Register r);
1000   void encode_heap_oop_not_null(Register dst, Register src);
1001   void decode_heap_oop_not_null(Register dst, Register src);
1002 
1003   void set_narrow_oop(Register dst, jobject obj);
1004 
1005   void decode_klass_not_null_for_aot(Register dst, Register src);

1015   void reinit_heapbase();
1016 
1017   DEBUG_ONLY(void verify_heapbase(const char* msg);)
1018 
1019   void push_CPU_state(bool save_vectors = false, bool use_sve = false,
1020                       int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1021   void pop_CPU_state(bool restore_vectors = false, bool use_sve = false,
1022                      int sve_vector_size_in_bytes = 0, int total_predicate_in_bytes = 0);
1023 
1024   void push_cont_fastpath(Register java_thread = rthread);
1025   void pop_cont_fastpath(Register java_thread = rthread);
1026 
1027   // Round up to a power of two
1028   void round_to(Register reg, int modulus);
1029 
1030   // java.lang.Math::round intrinsics
1031   void java_round_double(Register dst, FloatRegister src, FloatRegister ftmp);
1032   void java_round_float(Register dst, FloatRegister src, FloatRegister ftmp);
1033 
1034   // allocation
1035 
1036   // Object / value buffer allocation...
1037   // Allocate instance of klass, assumes klass initialized by caller
1038   // new_obj prefers to be rax
1039   // Kills t1 and t2, perserves klass, return allocation in new_obj (rsi on LP64)
1040   void allocate_instance(Register klass, Register new_obj,
1041                          Register t1, Register t2,
1042                          bool clear_fields, Label& alloc_failed);
1043 
1044   void tlab_allocate(
1045     Register obj,                      // result: pointer to object after successful allocation
1046     Register var_size_in_bytes,        // object size in bytes if unknown at compile time; invalid otherwise
1047     int      con_size_in_bytes,        // object size in bytes if   known at compile time
1048     Register t1,                       // temp register
1049     Register t2,                       // temp register
1050     Label&   slow_case                 // continuation point if fast allocation fails
1051   );
1052   void verify_tlab();
1053 
1054   void inline_layout_info(Register holder_klass, Register index, Register layout_info);
1055 
1056   // interface method calling
1057   void lookup_interface_method(Register recv_klass,
1058                                Register intf_klass,
1059                                RegisterOrConstant itable_index,
1060                                Register method_result,
1061                                Register scan_temp,
1062                                Label& no_such_interface,
1063                    bool return_method = true);
1064 
1065   void lookup_interface_method_stub(Register recv_klass,
1066                                     Register holder_klass,
1067                                     Register resolved_klass,
1068                                     Register method_result,
1069                                     Register temp_reg,
1070                                     Register temp_reg2,
1071                                     int itable_index,
1072                                     Label& L_no_such_interface);
1073 
1074   // virtual method calling
1075   // n.b. x86 allows RegisterOrConstant for vtable_index

1493   }                                                                     \
1494                                                                         \
1495   void INSN(Register Rd, Register Rn, Register Rm) {                    \
1496     Assembler::INSN(Rd, Rn, Rm);                                        \
1497   }                                                                     \
1498                                                                         \
1499   void INSN(Register Rd, Register Rn, Register Rm,                      \
1500            ext::operation option, int amount = 0) {                     \
1501     Assembler::INSN(Rd, Rn, Rm, option, amount);                        \
1502   }
1503 
1504   WRAP(adds, false) WRAP(addsw, true) WRAP(subs, false) WRAP(subsw, true)
1505 
1506   void add(Register Rd, Register Rn, RegisterOrConstant increment);
1507   void addw(Register Rd, Register Rn, RegisterOrConstant increment);
1508   void sub(Register Rd, Register Rn, RegisterOrConstant decrement);
1509   void subw(Register Rd, Register Rn, RegisterOrConstant decrement);
1510 
1511   void adrp(Register reg1, const Address &dest, uint64_t &byte_offset);
1512 
1513   void verified_entry(Compile* C, int sp_inc);
1514 
1515   // Inline type specific methods
1516   #include "asm/macroAssembler_common.hpp"
1517 
1518   int store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter = true);
1519   bool move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]);
1520   bool unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
1521                             VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
1522                             RegState reg_state[]);
1523   bool pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
1524                           VMRegPair* from, int from_count, int& from_index, VMReg to,
1525                           RegState reg_state[], Register val_array);
1526   int extend_stack_for_inline_args(int args_on_stack);
1527   void remove_frame(int initial_framesize, bool needs_stack_repair);
1528   VMReg spill_reg_for(VMReg reg);
1529   void save_stack_increment(int sp_inc, int frame_size);
1530 
1531   void tableswitch(Register index, jint lowbound, jint highbound,
1532                    Label &jumptable, Label &jumptable_end, int stride = 1) {
1533     adr(rscratch1, jumptable);
1534     subsw(rscratch2, index, lowbound);
1535     subsw(zr, rscratch2, highbound - lowbound);
1536     br(Assembler::HS, jumptable_end);
1537     add(rscratch1, rscratch1, rscratch2,
1538         ext::sxtw, exact_log2(stride * Assembler::instruction_size));
1539     br(rscratch1);
1540   }
1541 
1542   // Form an address from base + offset in Rd.  Rd may or may not
1543   // actually be used: you must use the Address that is returned.  It
1544   // is up to you to ensure that the shift provided matches the size
1545   // of your data.
1546   Address form_address(Register Rd, Register base, int64_t byte_offset, int shift);
1547 
1548   // Return true iff an address is within the 48-bit AArch64 address
1549   // space.
1550   bool is_valid_AArch64_address(address a) {

1585 #define ARRAYS_HASHCODE_REGISTERS \
1586   do {                      \
1587     assert(result == r0  && \
1588            ary    == r1  && \
1589            cnt    == r2  && \
1590            vdata0 == v3  && \
1591            vdata1 == v2  && \
1592            vdata2 == v1  && \
1593            vdata3 == v0  && \
1594            vmul0  == v4  && \
1595            vmul1  == v5  && \
1596            vmul2  == v6  && \
1597            vmul3  == v7  && \
1598            vpow   == v12 && \
1599            vpowm  == v13, "registers must match aarch64.ad"); \
1600   } while (0)
1601 
1602   void string_equals(Register a1, Register a2, Register result, Register cnt1);
1603 
1604   void fill_words(Register base, Register cnt, Register value);
1605   void fill_words(Register base, uint64_t cnt, Register value);
1606 
1607   address zero_words(Register base, uint64_t cnt);
1608   address zero_words(Register ptr, Register cnt);
1609   void zero_dcache_blocks(Register base, Register cnt);
1610 
1611   static const int zero_words_block_size;
1612 
1613   address byte_array_inflate(Register src, Register dst, Register len,
1614                              FloatRegister vtmp1, FloatRegister vtmp2,
1615                              FloatRegister vtmp3, Register tmp4);
1616 
1617   void char_array_compress(Register src, Register dst, Register len,
1618                            Register res,
1619                            FloatRegister vtmp0, FloatRegister vtmp1,
1620                            FloatRegister vtmp2, FloatRegister vtmp3,
1621                            FloatRegister vtmp4, FloatRegister vtmp5);
1622 
1623   void encode_iso_array(Register src, Register dst,
1624                         Register len, Register res, bool ascii,
1625                         FloatRegister vtmp0, FloatRegister vtmp1,
1626                         FloatRegister vtmp2, FloatRegister vtmp3,
< prev index next >