src/share/vm/runtime/stubRoutines.hpp

Wed, 11 Jun 2014 11:05:10 -0700

author
kvn
date
Wed, 11 Jun 2014 11:05:10 -0700
changeset 7027
b20a35eae442
parent 6457
94c202aa2646
child 7152
166d744df0de
permissions
-rw-r--r--

8035968: Leverage CPU Instructions to Improve SHA Performance on SPARC
Summary: Add C2 SHA intrinsics on SPARC
Reviewed-by: kvn, roland
Contributed-by: james.cheng@oracle.com

     1 /*
     2  * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
     3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
     4  *
     5  * This code is free software; you can redistribute it and/or modify it
     6  * under the terms of the GNU General Public License version 2 only, as
     7  * published by the Free Software Foundation.
     8  *
     9  * This code is distributed in the hope that it will be useful, but WITHOUT
    10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
    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  */
    25 #ifndef SHARE_VM_RUNTIME_STUBROUTINES_HPP
    26 #define SHARE_VM_RUNTIME_STUBROUTINES_HPP
    28 #include "code/codeBlob.hpp"
    29 #include "memory/allocation.hpp"
    30 #include "runtime/frame.hpp"
    31 #include "runtime/mutexLocker.hpp"
    32 #include "runtime/stubCodeGenerator.hpp"
    33 #include "utilities/top.hpp"
    34 #ifdef TARGET_ARCH_x86
    35 # include "nativeInst_x86.hpp"
    36 #endif
    37 #ifdef TARGET_ARCH_sparc
    38 # include "nativeInst_sparc.hpp"
    39 #endif
    40 #ifdef TARGET_ARCH_zero
    41 # include "nativeInst_zero.hpp"
    42 #endif
    43 #ifdef TARGET_ARCH_arm
    44 # include "nativeInst_arm.hpp"
    45 #endif
    46 #ifdef TARGET_ARCH_ppc
    47 # include "nativeInst_ppc.hpp"
    48 #endif
    50 // StubRoutines provides entry points to assembly routines used by
    51 // compiled code and the run-time system. Platform-specific entry
    52 // points are defined in the platform-specific inner class.
    53 //
    54 // Class scheme:
    55 //
    56 //    platform-independent               platform-dependent
    57 //
    58 //    stubRoutines.hpp  <-- included --  stubRoutines_<arch>.hpp
    59 //           ^                                  ^
    60 //           |                                  |
    61 //       implements                         implements
    62 //           |                                  |
    63 //           |                                  |
    64 //    stubRoutines.cpp                   stubRoutines_<arch>.cpp
    65 //    stubRoutines_<os_family>.cpp       stubGenerator_<arch>.cpp
    66 //    stubRoutines_<os_arch>.cpp
    67 //
    68 // Note 1: The important thing is a clean decoupling between stub
    69 //         entry points (interfacing to the whole vm; i.e., 1-to-n
    70 //         relationship) and stub generators (interfacing only to
    71 //         the entry points implementation; i.e., 1-to-1 relationship).
    72 //         This significantly simplifies changes in the generator
    73 //         structure since the rest of the vm is not affected.
    74 //
    75 // Note 2: stubGenerator_<arch>.cpp contains a minimal portion of
    76 //         machine-independent code; namely the generator calls of
    77 //         the generator functions that are used platform-independently.
    78 //         However, it comes with the advantage of having a 1-file
    79 //         implementation of the generator. It should be fairly easy
    80 //         to change, should it become a problem later.
    81 //
    82 // Scheme for adding a new entry point:
    83 //
    84 // 1. determine if it's a platform-dependent or independent entry point
    85 //    a) if platform independent: make subsequent changes in the independent files
    86 //    b) if platform   dependent: make subsequent changes in the   dependent files
    87 // 2. add a private instance variable holding the entry point address
    88 // 3. add a public accessor function to the instance variable
    89 // 4. implement the corresponding generator function in the platform-dependent
    90 //    stubGenerator_<arch>.cpp file and call the function in generate_all() of that file
    93 class StubRoutines: AllStatic {
    95  public:
    96   enum platform_independent_constants {
    97     max_size_of_parameters = 256                           // max. parameter size supported by megamorphic lookups
    98   };
   100   // Dependencies
   101   friend class StubGenerator;
   102 #ifdef TARGET_ARCH_MODEL_x86_32
   103 # include "stubRoutines_x86_32.hpp"
   104 #endif
   105 #ifdef TARGET_ARCH_MODEL_x86_64
   106 # include "stubRoutines_x86_64.hpp"
   107 #endif
   108 #ifdef TARGET_ARCH_MODEL_sparc
   109 # include "stubRoutines_sparc.hpp"
   110 #endif
   111 #ifdef TARGET_ARCH_MODEL_zero
   112 # include "stubRoutines_zero.hpp"
   113 #endif
   114 #ifdef TARGET_ARCH_MODEL_arm
   115 # include "stubRoutines_arm.hpp"
   116 #endif
   117 #ifdef TARGET_ARCH_MODEL_ppc_32
   118 # include "stubRoutines_ppc_32.hpp"
   119 #endif
   120 #ifdef TARGET_ARCH_MODEL_ppc_64
   121 # include "stubRoutines_ppc_64.hpp"
   122 #endif
   125   static jint    _verify_oop_count;
   126   static address _verify_oop_subroutine_entry;
   128   static address _call_stub_return_address;                // the return PC, when returning to a call stub
   129   static address _call_stub_entry;
   130   static address _forward_exception_entry;
   131   static address _catch_exception_entry;
   132   static address _throw_AbstractMethodError_entry;
   133   static address _throw_IncompatibleClassChangeError_entry;
   134   static address _throw_NullPointerException_at_call_entry;
   135   static address _throw_StackOverflowError_entry;
   136   static address _handler_for_unsafe_access_entry;
   138   static address _atomic_xchg_entry;
   139   static address _atomic_xchg_ptr_entry;
   140   static address _atomic_store_entry;
   141   static address _atomic_store_ptr_entry;
   142   static address _atomic_cmpxchg_entry;
   143   static address _atomic_cmpxchg_ptr_entry;
   144   static address _atomic_cmpxchg_long_entry;
   145   static address _atomic_add_entry;
   146   static address _atomic_add_ptr_entry;
   147   static address _fence_entry;
   148   static address _d2i_wrapper;
   149   static address _d2l_wrapper;
   151   static jint    _fpu_cntrl_wrd_std;
   152   static jint    _fpu_cntrl_wrd_24;
   153   static jint    _fpu_cntrl_wrd_64;
   154   static jint    _fpu_cntrl_wrd_trunc;
   155   static jint    _mxcsr_std;
   156   static jint    _fpu_subnormal_bias1[3];
   157   static jint    _fpu_subnormal_bias2[3];
   159   static BufferBlob* _code1;                               // code buffer for initial routines
   160   static BufferBlob* _code2;                               // code buffer for all other routines
   162   // Leaf routines which implement arraycopy and their addresses
   163   // arraycopy operands aligned on element type boundary
   164   static address _jbyte_arraycopy;
   165   static address _jshort_arraycopy;
   166   static address _jint_arraycopy;
   167   static address _jlong_arraycopy;
   168   static address _oop_arraycopy, _oop_arraycopy_uninit;
   169   static address _jbyte_disjoint_arraycopy;
   170   static address _jshort_disjoint_arraycopy;
   171   static address _jint_disjoint_arraycopy;
   172   static address _jlong_disjoint_arraycopy;
   173   static address _oop_disjoint_arraycopy, _oop_disjoint_arraycopy_uninit;
   175   // arraycopy operands aligned on zero'th element boundary
   176   // These are identical to the ones aligned aligned on an
   177   // element type boundary, except that they assume that both
   178   // source and destination are HeapWord aligned.
   179   static address _arrayof_jbyte_arraycopy;
   180   static address _arrayof_jshort_arraycopy;
   181   static address _arrayof_jint_arraycopy;
   182   static address _arrayof_jlong_arraycopy;
   183   static address _arrayof_oop_arraycopy, _arrayof_oop_arraycopy_uninit;
   184   static address _arrayof_jbyte_disjoint_arraycopy;
   185   static address _arrayof_jshort_disjoint_arraycopy;
   186   static address _arrayof_jint_disjoint_arraycopy;
   187   static address _arrayof_jlong_disjoint_arraycopy;
   188   static address _arrayof_oop_disjoint_arraycopy, _arrayof_oop_disjoint_arraycopy_uninit;
   190   // these are recommended but optional:
   191   static address _checkcast_arraycopy, _checkcast_arraycopy_uninit;
   192   static address _unsafe_arraycopy;
   193   static address _generic_arraycopy;
   195   static address _jbyte_fill;
   196   static address _jshort_fill;
   197   static address _jint_fill;
   198   static address _arrayof_jbyte_fill;
   199   static address _arrayof_jshort_fill;
   200   static address _arrayof_jint_fill;
   202   // zero heap space aligned to jlong (8 bytes)
   203   static address _zero_aligned_words;
   205   static address _aescrypt_encryptBlock;
   206   static address _aescrypt_decryptBlock;
   207   static address _cipherBlockChaining_encryptAESCrypt;
   208   static address _cipherBlockChaining_decryptAESCrypt;
   210   static address _sha1_implCompress;
   211   static address _sha1_implCompressMB;
   212   static address _sha256_implCompress;
   213   static address _sha256_implCompressMB;
   214   static address _sha512_implCompress;
   215   static address _sha512_implCompressMB;
   217   static address _updateBytesCRC32;
   218   static address _crc_table_adr;
   220   // These are versions of the java.lang.Math methods which perform
   221   // the same operations as the intrinsic version.  They are used for
   222   // constant folding in the compiler to ensure equivalence.  If the
   223   // intrinsic version returns the same result as the strict version
   224   // then they can be set to the appropriate function from
   225   // SharedRuntime.
   226   static double (*_intrinsic_log)(double);
   227   static double (*_intrinsic_log10)(double);
   228   static double (*_intrinsic_exp)(double);
   229   static double (*_intrinsic_pow)(double, double);
   230   static double (*_intrinsic_sin)(double);
   231   static double (*_intrinsic_cos)(double);
   232   static double (*_intrinsic_tan)(double);
   234   // Safefetch stubs.
   235   static address _safefetch32_entry;
   236   static address _safefetch32_fault_pc;
   237   static address _safefetch32_continuation_pc;
   238   static address _safefetchN_entry;
   239   static address _safefetchN_fault_pc;
   240   static address _safefetchN_continuation_pc;
   242  public:
   243   // Initialization/Testing
   244   static void    initialize1();                            // must happen before universe::genesis
   245   static void    initialize2();                            // must happen after  universe::genesis
   247   static bool is_stub_code(address addr)                   { return contains(addr); }
   249   static bool contains(address addr) {
   250     return
   251       (_code1 != NULL && _code1->blob_contains(addr)) ||
   252       (_code2 != NULL && _code2->blob_contains(addr)) ;
   253   }
   255   static CodeBlob* code1() { return _code1; }
   256   static CodeBlob* code2() { return _code2; }
   258   // Debugging
   259   static jint    verify_oop_count()                        { return _verify_oop_count; }
   260   static jint*   verify_oop_count_addr()                   { return &_verify_oop_count; }
   261   // a subroutine for debugging the GC
   262   static address verify_oop_subroutine_entry_address()    { return (address)&_verify_oop_subroutine_entry; }
   264   static address catch_exception_entry()                   { return _catch_exception_entry; }
   266   // Calls to Java
   267   typedef void (*CallStub)(
   268     address   link,
   269     intptr_t* result,
   270     BasicType result_type,
   271     Method* method,
   272     address   entry_point,
   273     intptr_t* parameters,
   274     int       size_of_parameters,
   275     TRAPS
   276   );
   278   static CallStub call_stub()                              { return CAST_TO_FN_PTR(CallStub, _call_stub_entry); }
   280   // Exceptions
   281   static address forward_exception_entry()                 { return _forward_exception_entry; }
   282   // Implicit exceptions
   283   static address throw_AbstractMethodError_entry()         { return _throw_AbstractMethodError_entry; }
   284   static address throw_IncompatibleClassChangeError_entry(){ return _throw_IncompatibleClassChangeError_entry; }
   285   static address throw_NullPointerException_at_call_entry(){ return _throw_NullPointerException_at_call_entry; }
   286   static address throw_StackOverflowError_entry()          { return _throw_StackOverflowError_entry; }
   288   // Exceptions during unsafe access - should throw Java exception rather
   289   // than crash.
   290   static address handler_for_unsafe_access()               { return _handler_for_unsafe_access_entry; }
   292   static address atomic_xchg_entry()                       { return _atomic_xchg_entry; }
   293   static address atomic_xchg_ptr_entry()                   { return _atomic_xchg_ptr_entry; }
   294   static address atomic_store_entry()                      { return _atomic_store_entry; }
   295   static address atomic_store_ptr_entry()                  { return _atomic_store_ptr_entry; }
   296   static address atomic_cmpxchg_entry()                    { return _atomic_cmpxchg_entry; }
   297   static address atomic_cmpxchg_ptr_entry()                { return _atomic_cmpxchg_ptr_entry; }
   298   static address atomic_cmpxchg_long_entry()               { return _atomic_cmpxchg_long_entry; }
   299   static address atomic_add_entry()                        { return _atomic_add_entry; }
   300   static address atomic_add_ptr_entry()                    { return _atomic_add_ptr_entry; }
   301   static address fence_entry()                             { return _fence_entry; }
   303   static address d2i_wrapper()                             { return _d2i_wrapper; }
   304   static address d2l_wrapper()                             { return _d2l_wrapper; }
   305   static jint    fpu_cntrl_wrd_std()                       { return _fpu_cntrl_wrd_std;   }
   306   static address addr_fpu_cntrl_wrd_std()                  { return (address)&_fpu_cntrl_wrd_std;   }
   307   static address addr_fpu_cntrl_wrd_24()                   { return (address)&_fpu_cntrl_wrd_24;   }
   308   static address addr_fpu_cntrl_wrd_64()                   { return (address)&_fpu_cntrl_wrd_64;   }
   309   static address addr_fpu_cntrl_wrd_trunc()                { return (address)&_fpu_cntrl_wrd_trunc; }
   310   static address addr_mxcsr_std()                          { return (address)&_mxcsr_std; }
   311   static address addr_fpu_subnormal_bias1()                { return (address)&_fpu_subnormal_bias1; }
   312   static address addr_fpu_subnormal_bias2()                { return (address)&_fpu_subnormal_bias2; }
   315   static address select_arraycopy_function(BasicType t, bool aligned, bool disjoint, const char* &name, bool dest_uninitialized);
   317   static address jbyte_arraycopy()  { return _jbyte_arraycopy; }
   318   static address jshort_arraycopy() { return _jshort_arraycopy; }
   319   static address jint_arraycopy()   { return _jint_arraycopy; }
   320   static address jlong_arraycopy()  { return _jlong_arraycopy; }
   321   static address oop_arraycopy(bool dest_uninitialized = false) {
   322     return dest_uninitialized ? _oop_arraycopy_uninit : _oop_arraycopy;
   323   }
   324   static address jbyte_disjoint_arraycopy()  { return _jbyte_disjoint_arraycopy; }
   325   static address jshort_disjoint_arraycopy() { return _jshort_disjoint_arraycopy; }
   326   static address jint_disjoint_arraycopy()   { return _jint_disjoint_arraycopy; }
   327   static address jlong_disjoint_arraycopy()  { return _jlong_disjoint_arraycopy; }
   328   static address oop_disjoint_arraycopy(bool dest_uninitialized = false) {
   329     return dest_uninitialized ?  _oop_disjoint_arraycopy_uninit : _oop_disjoint_arraycopy;
   330   }
   332   static address arrayof_jbyte_arraycopy()  { return _arrayof_jbyte_arraycopy; }
   333   static address arrayof_jshort_arraycopy() { return _arrayof_jshort_arraycopy; }
   334   static address arrayof_jint_arraycopy()   { return _arrayof_jint_arraycopy; }
   335   static address arrayof_jlong_arraycopy()  { return _arrayof_jlong_arraycopy; }
   336   static address arrayof_oop_arraycopy(bool dest_uninitialized = false) {
   337     return dest_uninitialized ? _arrayof_oop_arraycopy_uninit : _arrayof_oop_arraycopy;
   338   }
   340   static address arrayof_jbyte_disjoint_arraycopy()  { return _arrayof_jbyte_disjoint_arraycopy; }
   341   static address arrayof_jshort_disjoint_arraycopy() { return _arrayof_jshort_disjoint_arraycopy; }
   342   static address arrayof_jint_disjoint_arraycopy()   { return _arrayof_jint_disjoint_arraycopy; }
   343   static address arrayof_jlong_disjoint_arraycopy()  { return _arrayof_jlong_disjoint_arraycopy; }
   344   static address arrayof_oop_disjoint_arraycopy(bool dest_uninitialized = false) {
   345     return dest_uninitialized ? _arrayof_oop_disjoint_arraycopy_uninit : _arrayof_oop_disjoint_arraycopy;
   346   }
   348   static address checkcast_arraycopy(bool dest_uninitialized = false) {
   349     return dest_uninitialized ? _checkcast_arraycopy_uninit : _checkcast_arraycopy;
   350   }
   351   static address unsafe_arraycopy()        { return _unsafe_arraycopy; }
   352   static address generic_arraycopy()       { return _generic_arraycopy; }
   354   static address jbyte_fill()          { return _jbyte_fill; }
   355   static address jshort_fill()         { return _jshort_fill; }
   356   static address jint_fill()           { return _jint_fill; }
   357   static address arrayof_jbyte_fill()  { return _arrayof_jbyte_fill; }
   358   static address arrayof_jshort_fill() { return _arrayof_jshort_fill; }
   359   static address arrayof_jint_fill()   { return _arrayof_jint_fill; }
   361   static address aescrypt_encryptBlock()                { return _aescrypt_encryptBlock; }
   362   static address aescrypt_decryptBlock()                { return _aescrypt_decryptBlock; }
   363   static address cipherBlockChaining_encryptAESCrypt()  { return _cipherBlockChaining_encryptAESCrypt; }
   364   static address cipherBlockChaining_decryptAESCrypt()  { return _cipherBlockChaining_decryptAESCrypt; }
   366   static address sha1_implCompress()     { return _sha1_implCompress; }
   367   static address sha1_implCompressMB()   { return _sha1_implCompressMB; }
   368   static address sha256_implCompress()   { return _sha256_implCompress; }
   369   static address sha256_implCompressMB() { return _sha256_implCompressMB; }
   370   static address sha512_implCompress()   { return _sha512_implCompress; }
   371   static address sha512_implCompressMB() { return _sha512_implCompressMB; }
   373   static address updateBytesCRC32()    { return _updateBytesCRC32; }
   374   static address crc_table_addr()      { return _crc_table_adr; }
   376   static address select_fill_function(BasicType t, bool aligned, const char* &name);
   378   static address zero_aligned_words()   { return _zero_aligned_words; }
   380   static double  intrinsic_log(double d) {
   381     assert(_intrinsic_log != NULL, "must be defined");
   382     return _intrinsic_log(d);
   383   }
   384   static double  intrinsic_log10(double d) {
   385     assert(_intrinsic_log != NULL, "must be defined");
   386     return _intrinsic_log10(d);
   387   }
   388   static double  intrinsic_exp(double d) {
   389     assert(_intrinsic_exp != NULL, "must be defined");
   390     return _intrinsic_exp(d);
   391   }
   392   static double  intrinsic_pow(double d, double d2) {
   393     assert(_intrinsic_pow != NULL, "must be defined");
   394     return _intrinsic_pow(d, d2);
   395   }
   396   static double  intrinsic_sin(double d) {
   397     assert(_intrinsic_sin != NULL, "must be defined");
   398     return _intrinsic_sin(d);
   399   }
   400   static double  intrinsic_cos(double d) {
   401     assert(_intrinsic_cos != NULL, "must be defined");
   402     return _intrinsic_cos(d);
   403   }
   404   static double  intrinsic_tan(double d) {
   405     assert(_intrinsic_tan != NULL, "must be defined");
   406     return _intrinsic_tan(d);
   407   }
   409   //
   410   // Safefetch stub support
   411   //
   413   typedef int      (*SafeFetch32Stub)(int*      adr, int      errValue);
   414   typedef intptr_t (*SafeFetchNStub) (intptr_t* adr, intptr_t errValue);
   416   static SafeFetch32Stub SafeFetch32_stub() { return CAST_TO_FN_PTR(SafeFetch32Stub, _safefetch32_entry); }
   417   static SafeFetchNStub  SafeFetchN_stub()  { return CAST_TO_FN_PTR(SafeFetchNStub,  _safefetchN_entry); }
   419   static bool is_safefetch_fault(address pc) {
   420     return pc != NULL &&
   421           (pc == _safefetch32_fault_pc ||
   422            pc == _safefetchN_fault_pc);
   423   }
   425   static address continuation_for_safefetch_fault(address pc) {
   426     assert(_safefetch32_continuation_pc != NULL &&
   427            _safefetchN_continuation_pc  != NULL,
   428            "not initialized");
   430     if (pc == _safefetch32_fault_pc) return _safefetch32_continuation_pc;
   431     if (pc == _safefetchN_fault_pc)  return _safefetchN_continuation_pc;
   433     ShouldNotReachHere();
   434     return NULL;
   435   }
   437   //
   438   // Default versions of the above arraycopy functions for platforms which do
   439   // not have specialized versions
   440   //
   441   static void jbyte_copy     (jbyte*  src, jbyte*  dest, size_t count);
   442   static void jshort_copy    (jshort* src, jshort* dest, size_t count);
   443   static void jint_copy      (jint*   src, jint*   dest, size_t count);
   444   static void jlong_copy     (jlong*  src, jlong*  dest, size_t count);
   445   static void oop_copy       (oop*    src, oop*    dest, size_t count);
   446   static void oop_copy_uninit(oop*    src, oop*    dest, size_t count);
   448   static void arrayof_jbyte_copy     (HeapWord* src, HeapWord* dest, size_t count);
   449   static void arrayof_jshort_copy    (HeapWord* src, HeapWord* dest, size_t count);
   450   static void arrayof_jint_copy      (HeapWord* src, HeapWord* dest, size_t count);
   451   static void arrayof_jlong_copy     (HeapWord* src, HeapWord* dest, size_t count);
   452   static void arrayof_oop_copy       (HeapWord* src, HeapWord* dest, size_t count);
   453   static void arrayof_oop_copy_uninit(HeapWord* src, HeapWord* dest, size_t count);
   454 };
   456 // Safefetch allows to load a value from a location that's not known
   457 // to be valid. If the load causes a fault, the error value is returned.
   458 inline int SafeFetch32(int* adr, int errValue) {
   459   assert(StubRoutines::SafeFetch32_stub(), "stub not yet generated");
   460   return StubRoutines::SafeFetch32_stub()(adr, errValue);
   461 }
   462 inline intptr_t SafeFetchN(intptr_t* adr, intptr_t errValue) {
   463   assert(StubRoutines::SafeFetchN_stub(), "stub not yet generated");
   464   return StubRoutines::SafeFetchN_stub()(adr, errValue);
   465 }
   467 #endif // SHARE_VM_RUNTIME_STUBROUTINES_HPP

mercurial