# HG changeset patch # User kvn # Date 1248451260 25200 # Node ID 433f394ab5093280be2876d6c24da3515100eb72 # Parent f0a1cbbaf3c02fbfb196929a93f12c3310be7796# Parent ea3f9723b5cff542eb5fa1e2ec2adb23e011e3d4 Merge diff -r f0a1cbbaf3c0 -r 433f394ab509 src/cpu/x86/vm/sharedRuntime_x86_64.cpp --- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -1302,22 +1302,19 @@ const Register ic_reg = rax; const Register receiver = j_rarg0; - const Register tmp = rdx; Label ok; Label exception_pending; + assert_different_registers(ic_reg, receiver, rscratch1); __ verify_oop(receiver); - __ push(tmp); // spill (any other registers free here???) - __ load_klass(tmp, receiver); - __ cmpq(ic_reg, tmp); + __ load_klass(rscratch1, receiver); + __ cmpq(ic_reg, rscratch1); __ jcc(Assembler::equal, ok); - __ pop(tmp); __ jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub())); __ bind(ok); - __ pop(tmp); // Verified entry point must be aligned __ align(8); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/adlc/formssel.cpp --- a/src/share/vm/adlc/formssel.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/adlc/formssel.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -420,6 +420,13 @@ return _matrule->is_ideal_load(); } +// Return 'true' if this instruction matches an ideal 'LoadKlass' node +bool InstructForm::skip_antidep_check() const { + if( _matrule == NULL ) return false; + + return _matrule->skip_antidep_check(); +} + // Return 'true' if this instruction matches an ideal 'Load?' node Form::DataType InstructForm::is_ideal_store() const { if( _matrule == NULL ) return Form::none; @@ -567,6 +574,8 @@ // loads from memory, so must check for anti-dependence bool InstructForm::needs_anti_dependence_check(FormDict &globals) const { + if ( skip_antidep_check() ) return false; + // Machine independent loads must be checked for anti-dependences if( is_ideal_load() != Form::none ) return true; @@ -3957,6 +3966,28 @@ } +bool MatchRule::skip_antidep_check() const { + // Some loads operate on what is effectively immutable memory so we + // should skip the anti dep computations. For some of these nodes + // the rewritable field keeps the anti dep logic from triggering but + // for certain kinds of LoadKlass it does not since they are + // actually reading memory which could be rewritten by the runtime, + // though never by generated code. This disables it uniformly for + // the nodes that behave like this: LoadKlass, LoadNKlass and + // LoadRange. + if ( _opType && (strcmp(_opType,"Set") == 0) && _rChild ) { + const char *opType = _rChild->_opType; + if (strcmp("LoadKlass", opType) == 0 || + strcmp("LoadNKlass", opType) == 0 || + strcmp("LoadRange", opType) == 0) { + return true; + } + } + + return false; +} + + Form::DataType MatchRule::is_ideal_store() const { Form::DataType ideal_store = Form::none; diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/adlc/formssel.hpp --- a/src/share/vm/adlc/formssel.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/adlc/formssel.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -158,6 +158,9 @@ virtual Form::CallType is_ideal_call() const; // matches ideal 'Call' virtual Form::DataType is_ideal_load() const; // node matches ideal 'LoadXNode' + // Should antidep checks be disabled for this Instruct + // See definition of MatchRule::skip_antidep_check + bool skip_antidep_check() const; virtual Form::DataType is_ideal_store() const;// node matches ideal 'StoreXNode' bool is_ideal_mem() const { return is_ideal_load() != Form::none || is_ideal_store() != Form::none; } virtual uint two_address(FormDict &globals); // output reg must match input reg @@ -1003,6 +1006,9 @@ bool is_ideal_loopEnd() const; // node matches ideal 'LoopEnd' bool is_ideal_bool() const; // node matches ideal 'Bool' Form::DataType is_ideal_load() const;// node matches ideal 'LoadXNode' + // Should antidep checks be disabled for this rule + // See definition of MatchRule::skip_antidep_check + bool skip_antidep_check() const; Form::DataType is_ideal_store() const;// node matches ideal 'StoreXNode' // Check if 'mRule2' is a cisc-spill variant of this MatchRule diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/classfile/classFileParser.cpp --- a/src/share/vm/classfile/classFileParser.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/classfile/classFileParser.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -3231,6 +3231,16 @@ this_klass->set_minor_version(minor_version); this_klass->set_major_version(major_version); + // Set up methodOop::intrinsic_id as soon as we know the names of methods. + // (We used to do this lazily, but now we query it in Rewriter, + // which is eagerly done for every method, so we might as well do it now, + // when everything is fresh in memory.) + if (methodOopDesc::klass_id_for_intrinsics(this_klass->as_klassOop()) != vmSymbols::NO_SID) { + for (int j = 0; j < methods->length(); j++) { + ((methodOop)methods->obj_at(j))->init_intrinsic_id(); + } + } + if (cached_class_file_bytes != NULL) { // JVMTI: we have an instanceKlass now, tell it about the cached bytes this_klass->set_cached_class_file(cached_class_file_bytes, diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/classfile/vmSymbols.hpp --- a/src/share/vm/classfile/vmSymbols.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/classfile/vmSymbols.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -513,9 +513,6 @@ // // for Emacs: (let ((c-backslash-column 120) (c-backslash-max-column 120)) (c-backslash-region (point) (point-max) nil t)) #define VM_INTRINSICS_DO(do_intrinsic, do_class, do_name, do_signature, do_alias) \ - do_intrinsic(_Object_init, java_lang_Object, object_initializer_name, void_method_signature, F_R) \ - /* (symbol object_initializer_name defined above) */ \ - \ do_intrinsic(_hashCode, java_lang_Object, hashCode_name, void_int_signature, F_R) \ do_name( hashCode_name, "hashCode") \ do_intrinsic(_getClass, java_lang_Object, getClass_name, void_class_signature, F_R) \ @@ -635,9 +632,6 @@ do_intrinsic(_equalsC, java_util_Arrays, equals_name, equalsC_signature, F_S) \ do_signature(equalsC_signature, "([C[C)Z") \ \ - do_intrinsic(_invoke, java_lang_reflect_Method, invoke_name, object_array_object_object_signature, F_R) \ - /* (symbols invoke_name and invoke_signature defined above) */ \ - \ do_intrinsic(_compareTo, java_lang_String, compareTo_name, string_int_signature, F_R) \ do_name( compareTo_name, "compareTo") \ do_intrinsic(_indexOf, java_lang_String, indexOf_name, string_int_signature, F_R) \ @@ -656,8 +650,6 @@ do_name( attemptUpdate_name, "attemptUpdate") \ do_signature(attemptUpdate_signature, "(JJ)Z") \ \ - do_intrinsic(_fillInStackTrace, java_lang_Throwable, fillInStackTrace_name, void_throwable_signature, F_RNY) \ - \ /* support for sun.misc.Unsafe */ \ do_class(sun_misc_Unsafe, "sun/misc/Unsafe") \ \ @@ -819,10 +811,22 @@ do_name( prefetchReadStatic_name, "prefetchReadStatic") \ do_intrinsic(_prefetchWriteStatic, sun_misc_Unsafe, prefetchWriteStatic_name, prefetch_signature, F_SN) \ do_name( prefetchWriteStatic_name, "prefetchWriteStatic") \ + /*== LAST_COMPILER_INLINE*/ \ + /*the compiler does have special inlining code for these; bytecode inline is just fine */ \ + \ + do_intrinsic(_fillInStackTrace, java_lang_Throwable, fillInStackTrace_name, void_throwable_signature, F_RNY) \ + \ + do_intrinsic(_Object_init, java_lang_Object, object_initializer_name, void_method_signature, F_R) \ + /* (symbol object_initializer_name defined above) */ \ + \ + do_intrinsic(_invoke, java_lang_reflect_Method, invoke_name, object_array_object_object_signature, F_R) \ + /* (symbols invoke_name and invoke_signature defined above) */ \ + \ /*end*/ + // Class vmSymbols class vmSymbols: AllStatic { @@ -935,6 +939,7 @@ #undef VM_INTRINSIC_ENUM ID_LIMIT, + LAST_COMPILER_INLINE = _prefetchWriteStatic, FIRST_ID = _none + 1 }; @@ -972,4 +977,7 @@ static Flags flags_for(ID id); static const char* short_name_as_C_string(ID id, char* buf, int size); + + // Access to intrinsic methods: + static methodOop method_for(ID id); }; diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/compiler/oopMap.cpp --- a/src/share/vm/compiler/oopMap.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/compiler/oopMap.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -379,7 +379,15 @@ if ( loc != NULL ) { oop *base_loc = fr->oopmapreg_to_location(omv.content_reg(), reg_map); oop *derived_loc = loc; - derived_oop_fn(base_loc, derived_loc); + oop val = *base_loc; + if (val == (oop)NULL || Universe::is_narrow_oop_base(val)) { + // Ignore NULL oops and decoded NULL narrow oops which + // equal to Universe::narrow_oop_base when a narrow oop + // implicit null check is used in compiled code. + // The narrow_oop_base could be NULL or be the address + // of the page below heap depending on compressed oops mode. + } else + derived_oop_fn(base_loc, derived_loc); } oms.next(); } while (!oms.is_done()); @@ -394,6 +402,15 @@ oop* loc = fr->oopmapreg_to_location(omv.reg(),reg_map); if ( loc != NULL ) { if ( omv.type() == OopMapValue::oop_value ) { + oop val = *loc; + if (val == (oop)NULL || Universe::is_narrow_oop_base(val)) { + // Ignore NULL oops and decoded NULL narrow oops which + // equal to Universe::narrow_oop_base when a narrow oop + // implicit null check is used in compiled code. + // The narrow_oop_base could be NULL or be the address + // of the page below heap depending on compressed oops mode. + continue; + } #ifdef ASSERT if ((((uintptr_t)loc & (sizeof(*loc)-1)) != 0) || !Universe::heap()->is_in_or_null(*loc)) { @@ -410,6 +427,8 @@ #endif // ASSERT oop_fn->do_oop(loc); } else if ( omv.type() == OopMapValue::value_value ) { + assert((*loc) == (oop)NULL || !Universe::is_narrow_oop_base(*loc), + "found invalid value pointer"); value_fn->do_oop(loc); } else if ( omv.type() == OopMapValue::narrowoop_value ) { narrowOop *nl = (narrowOop*)loc; diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/compiler/oopMap.hpp --- a/src/share/vm/compiler/oopMap.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/compiler/oopMap.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -233,6 +233,10 @@ int heap_size() const; void copy_to(address addr); + // Methods oops_do() and all_do() filter out NULL oops and + // oop == Universe::narrow_oop_base() before passing oops + // to closures. + // Iterates through frame for a compiled method static void oops_do (const frame* fr, const RegisterMap* reg_map, OopClosure* f); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/interpreter/rewriter.cpp --- a/src/share/vm/interpreter/rewriter.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/interpreter/rewriter.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -273,6 +273,7 @@ compute_index_maps(); if (RegisterFinalizersAtInit && _klass->name() == vmSymbols::java_lang_Object()) { + bool did_rewrite = false; int i = _methods->length(); while (i-- > 0) { methodOop method = (methodOop)_methods->obj_at(i); @@ -281,9 +282,11 @@ // object for finalization if needed. methodHandle m(THREAD, method); rewrite_Object_init(m, CHECK); + did_rewrite = true; break; } } + assert(did_rewrite, "must find Object:: to rewrite it"); } // rewrite methods, in two passes diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/memory/universe.hpp --- a/src/share/vm/memory/universe.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/memory/universe.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -343,6 +343,7 @@ // For UseCompressedOops static address* narrow_oop_base_addr() { return &_narrow_oop._base; } static address narrow_oop_base() { return _narrow_oop._base; } + static bool is_narrow_oop_base(void* addr) { return (narrow_oop_base() == (address)addr); } static int narrow_oop_shift() { return _narrow_oop._shift; } static void set_narrow_oop_base(address base) { _narrow_oop._base = base; } static void set_narrow_oop_shift(int shift) { _narrow_oop._shift = shift; } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/oops/methodKlass.cpp --- a/src/share/vm/oops/methodKlass.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/oops/methodKlass.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -68,7 +68,7 @@ m->set_constants(NULL); m->set_max_stack(0); m->set_max_locals(0); - m->clear_intrinsic_id_cache(); + m->set_intrinsic_id(vmIntrinsics::_none); m->set_method_data(NULL); m->set_interpreter_throwout_count(0); m->set_vtable_index(methodOopDesc::garbage_vtable_index); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/oops/methodOop.cpp --- a/src/share/vm/oops/methodOop.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/oops/methodOop.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -962,26 +962,39 @@ return newm; } -vmIntrinsics::ID methodOopDesc::compute_intrinsic_id() const { - assert(vmIntrinsics::_none == 0, "correct coding of default case"); - const uintptr_t max_cache_uint = right_n_bits((int)(sizeof(_intrinsic_id_cache) * BitsPerByte)); - assert((uintptr_t)vmIntrinsics::ID_LIMIT <= max_cache_uint, "else fix cache size"); +vmSymbols::SID methodOopDesc::klass_id_for_intrinsics(klassOop holder) { // if loader is not the default loader (i.e., != NULL), we can't know the intrinsics // because we are not loading from core libraries - if (instanceKlass::cast(method_holder())->class_loader() != NULL) return vmIntrinsics::_none; + if (instanceKlass::cast(holder)->class_loader() != NULL) + return vmSymbols::NO_SID; // regardless of name, no intrinsics here // see if the klass name is well-known: - symbolOop klass_name = instanceKlass::cast(method_holder())->name(); - vmSymbols::SID klass_id = vmSymbols::find_sid(klass_name); - if (klass_id == vmSymbols::NO_SID) return vmIntrinsics::_none; + symbolOop klass_name = instanceKlass::cast(holder)->name(); + return vmSymbols::find_sid(klass_name); +} + +void methodOopDesc::init_intrinsic_id() { + assert(_intrinsic_id == vmIntrinsics::_none, "do this just once"); + const uintptr_t max_id_uint = right_n_bits((int)(sizeof(_intrinsic_id) * BitsPerByte)); + assert((uintptr_t)vmIntrinsics::ID_LIMIT <= max_id_uint, "else fix size"); + + // the klass name is well-known: + vmSymbols::SID klass_id = klass_id_for_intrinsics(method_holder()); + assert(klass_id != vmSymbols::NO_SID, "caller responsibility"); // ditto for method and signature: vmSymbols::SID name_id = vmSymbols::find_sid(name()); - if (name_id == vmSymbols::NO_SID) return vmIntrinsics::_none; + if (name_id == vmSymbols::NO_SID) return; vmSymbols::SID sig_id = vmSymbols::find_sid(signature()); - if (sig_id == vmSymbols::NO_SID) return vmIntrinsics::_none; + if (sig_id == vmSymbols::NO_SID) return; jshort flags = access_flags().as_short(); + vmIntrinsics::ID id = vmIntrinsics::find_id(klass_id, name_id, sig_id, flags); + if (id != vmIntrinsics::_none) { + set_intrinsic_id(id); + return; + } + // A few slightly irregular cases: switch (klass_id) { case vmSymbols::VM_SYMBOL_ENUM_NAME(java_lang_StrictMath): @@ -992,15 +1005,18 @@ case vmSymbols::VM_SYMBOL_ENUM_NAME(sqrt_name): // pretend it is the corresponding method in the non-strict class: klass_id = vmSymbols::VM_SYMBOL_ENUM_NAME(java_lang_Math); + id = vmIntrinsics::find_id(klass_id, name_id, sig_id, flags); break; } } - // return intrinsic id if any - return vmIntrinsics::find_id(klass_id, name_id, sig_id, flags); + if (id != vmIntrinsics::_none) { + // Set up its iid. It is an alias method. + set_intrinsic_id(id); + return; + } } - // These two methods are static since a GC may move the methodOopDesc bool methodOopDesc::load_signature_classes(methodHandle m, TRAPS) { bool sig_is_loaded = true; diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/oops/methodOop.hpp --- a/src/share/vm/oops/methodOop.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/oops/methodOop.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -104,7 +104,7 @@ u2 _max_stack; // Maximum number of entries on the expression stack u2 _max_locals; // Number of local variables used by this method u2 _size_of_parameters; // size of the parameter block (receiver + arguments) in words - u1 _intrinsic_id_cache; // Cache for intrinsic_id; 0 or 1+vmInt::ID + u1 _intrinsic_id; // vmSymbols::intrinsic_id (0 == _none) u1 _highest_tier_compile; // Highest compile level this method has ever seen. u2 _interpreter_throwout_count; // Count of times method was exited via exception while interpreting u2 _number_of_breakpoints; // fullspeed debugging support @@ -224,8 +224,6 @@ int highest_tier_compile() { return _highest_tier_compile;} void set_highest_tier_compile(int level) { _highest_tier_compile = level;} - void clear_intrinsic_id_cache() { _intrinsic_id_cache = 0; } - // Count of times method was exited via exception while interpreting void interpreter_throwout_increment() { if (_interpreter_throwout_count < 65534) { @@ -571,18 +569,12 @@ void set_cached_itable_index(int index) { instanceKlass::cast(method_holder())->set_cached_itable_index(method_idnum(), index); } // Support for inlining of intrinsic methods - vmIntrinsics::ID intrinsic_id() const { // returns zero if not an intrinsic - const u1& cache = _intrinsic_id_cache; - if (cache != 0) { - return (vmIntrinsics::ID)(cache - 1); - } else { - vmIntrinsics::ID id = compute_intrinsic_id(); - *(u1*)&cache = ((u1) id) + 1; // force the cache to be non-const - vmIntrinsics::verify_method(id, (methodOop) this); - assert((vmIntrinsics::ID)(cache - 1) == id, "proper conversion"); - return id; - } - } + vmIntrinsics::ID intrinsic_id() const { return (vmIntrinsics::ID) _intrinsic_id; } + void set_intrinsic_id(vmIntrinsics::ID id) { _intrinsic_id = (u1) id; } + + // Helper routines for intrinsic_id() and vmIntrinsics::method(). + void init_intrinsic_id(); // updates from _none if a match + static vmSymbols::SID klass_id_for_intrinsics(klassOop holder); // On-stack replacement support bool has_osr_nmethod() { return instanceKlass::cast(method_holder())->lookup_osr_nmethod(this, InvocationEntryBci) != NULL; } @@ -635,9 +627,6 @@ void set_size_of_parameters(int size) { _size_of_parameters = size; } private: - // Helper routine for intrinsic_id(). - vmIntrinsics::ID compute_intrinsic_id() const; - // Inlined elements address* native_function_addr() const { assert(is_native(), "must be native"); return (address*) (this+1); } address* signature_handler_addr() const { return native_function_addr() + 1; } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/cfgnode.cpp --- a/src/share/vm/opto/cfgnode.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/cfgnode.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -1789,15 +1789,19 @@ #ifdef _LP64 // Push DecodeN down through phi. // The rest of phi graph will transform by split EncodeP node though phis up. - if (UseNewCode && UseCompressedOops && can_reshape && progress == NULL) { + if (UseCompressedOops && can_reshape && progress == NULL) { bool may_push = true; bool has_decodeN = false; Node* in_decodeN = NULL; for (uint i=1; iis_DecodeN() && ii->bottom_type() == bottom_type()) { - has_decodeN = true; - in_decodeN = ii->in(1); + // Note: in_decodeN is used only to define the type of new phi. + // Find a non dead path otherwise phi type will be wrong. + if (ii->in(1)->bottom_type() != Type::TOP) { + has_decodeN = true; + in_decodeN = ii->in(1); + } } else if (!ii->is_Phi()) { may_push = false; } @@ -1805,7 +1809,6 @@ if (has_decodeN && may_push) { PhaseIterGVN *igvn = phase->is_IterGVN(); - // Note: in_decodeN is used only to define the type of new phi here. PhiNode *new_phi = PhiNode::make_blank(in(0), in_decodeN); uint orig_cnt = req(); for (uint i=1; iintrinsic_id() != vmIntrinsics::_none) { + if (m->intrinsic_id() != vmIntrinsics::_none && + m->intrinsic_id() <= vmIntrinsics::LAST_COMPILER_INLINE) { CallGenerator* cg = make_vm_intrinsic(m, is_virtual); if (cg != NULL) { // Save it for next time: @@ -440,6 +441,8 @@ _orig_pc_slot_offset_in_bytes(0), _node_bundling_limit(0), _node_bundling_base(NULL), + _java_calls(0), + _inner_loops(0), #ifndef PRODUCT _trace_opto_output(TraceOptoOutput || method()->has_option("TraceOptoOutput")), _printer(IdealGraphPrinter::printer()), @@ -710,6 +713,8 @@ _code_buffer("Compile::Fill_buffer"), _node_bundling_limit(0), _node_bundling_base(NULL), + _java_calls(0), + _inner_loops(0), #ifndef PRODUCT _trace_opto_output(TraceOptoOutput), _printer(NULL), @@ -1850,22 +1855,26 @@ int _float_count; // count float ops requiring 24-bit precision int _double_count; // count double ops requiring more precision int _java_call_count; // count non-inlined 'java' calls + int _inner_loop_count; // count loops which need alignment VectorSet _visited; // Visitation flags Node_List _tests; // Set of IfNodes & PCTableNodes Final_Reshape_Counts() : - _call_count(0), _float_count(0), _double_count(0), _java_call_count(0), + _call_count(0), _float_count(0), _double_count(0), + _java_call_count(0), _inner_loop_count(0), _visited( Thread::current()->resource_area() ) { } void inc_call_count () { _call_count ++; } void inc_float_count () { _float_count ++; } void inc_double_count() { _double_count++; } void inc_java_call_count() { _java_call_count++; } + void inc_inner_loop_count() { _inner_loop_count++; } int get_call_count () const { return _call_count ; } int get_float_count () const { return _float_count ; } int get_double_count() const { return _double_count; } int get_java_call_count() const { return _java_call_count; } + int get_inner_loop_count() const { return _inner_loop_count; } }; static bool oop_offset_is_sane(const TypeInstPtr* tp) { @@ -1877,7 +1886,7 @@ //------------------------------final_graph_reshaping_impl---------------------- // Implement items 1-5 from final_graph_reshaping below. -static void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &fpu ) { +static void final_graph_reshaping_impl( Node *n, Final_Reshape_Counts &frc ) { if ( n->outcnt() == 0 ) return; // dead node uint nop = n->Opcode(); @@ -1919,13 +1928,13 @@ case Op_CmpF: case Op_CmpF3: // case Op_ConvL2F: // longs are split into 32-bit halves - fpu.inc_float_count(); + frc.inc_float_count(); break; case Op_ConvF2D: case Op_ConvD2F: - fpu.inc_float_count(); - fpu.inc_double_count(); + frc.inc_float_count(); + frc.inc_double_count(); break; // Count all double operations that may use FPU @@ -1942,7 +1951,7 @@ case Op_ConD: case Op_CmpD: case Op_CmpD3: - fpu.inc_double_count(); + frc.inc_double_count(); break; case Op_Opaque1: // Remove Opaque Nodes before matching case Op_Opaque2: // Remove Opaque Nodes before matching @@ -1951,7 +1960,7 @@ case Op_CallStaticJava: case Op_CallJava: case Op_CallDynamicJava: - fpu.inc_java_call_count(); // Count java call site; + frc.inc_java_call_count(); // Count java call site; case Op_CallRuntime: case Op_CallLeaf: case Op_CallLeafNoFP: { @@ -1962,7 +1971,7 @@ // uncommon_trap, _complete_monitor_locking, _complete_monitor_unlocking, // _new_Java, _new_typeArray, _new_objArray, _rethrow_Java, ... if( !call->is_CallStaticJava() || !call->as_CallStaticJava()->_name ) { - fpu.inc_call_count(); // Count the call site + frc.inc_call_count(); // Count the call site } else { // See if uncommon argument is shared Node *n = call->in(TypeFunc::Parms); int nop = n->Opcode(); @@ -1983,11 +1992,11 @@ case Op_StoreD: case Op_LoadD: case Op_LoadD_unaligned: - fpu.inc_double_count(); + frc.inc_double_count(); goto handle_mem; case Op_StoreF: case Op_LoadF: - fpu.inc_float_count(); + frc.inc_float_count(); goto handle_mem; case Op_StoreB: @@ -2324,6 +2333,12 @@ n->subsume_by(btp); } break; + case Op_Loop: + case Op_CountedLoop: + if (n->as_Loop()->is_inner_loop()) { + frc.inc_inner_loop_count(); + } + break; default: assert( !n->is_Call(), "" ); assert( !n->is_Mem(), "" ); @@ -2332,17 +2347,17 @@ // Collect CFG split points if (n->is_MultiBranch()) - fpu._tests.push(n); + frc._tests.push(n); } //------------------------------final_graph_reshaping_walk--------------------- // Replacing Opaque nodes with their input in final_graph_reshaping_impl(), // requires that the walk visits a node's inputs before visiting the node. -static void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &fpu ) { +static void final_graph_reshaping_walk( Node_Stack &nstack, Node *root, Final_Reshape_Counts &frc ) { ResourceArea *area = Thread::current()->resource_area(); Unique_Node_List sfpt(area); - fpu._visited.set(root->_idx); // first, mark node as visited + frc._visited.set(root->_idx); // first, mark node as visited uint cnt = root->req(); Node *n = root; uint i = 0; @@ -2351,7 +2366,7 @@ // Place all non-visited non-null inputs onto stack Node* m = n->in(i); ++i; - if (m != NULL && !fpu._visited.test_set(m->_idx)) { + if (m != NULL && !frc._visited.test_set(m->_idx)) { if (m->is_SafePoint() && m->as_SafePoint()->jvms() != NULL) sfpt.push(m); cnt = m->req(); @@ -2361,7 +2376,7 @@ } } else { // Now do post-visit work - final_graph_reshaping_impl( n, fpu ); + final_graph_reshaping_impl( n, frc ); if (nstack.is_empty()) break; // finished n = nstack.node(); // Get node from stack @@ -2442,16 +2457,16 @@ return true; } - Final_Reshape_Counts fpu; + Final_Reshape_Counts frc; // Visit everybody reachable! // Allocate stack of size C->unique()/2 to avoid frequent realloc Node_Stack nstack(unique() >> 1); - final_graph_reshaping_walk(nstack, root(), fpu); + final_graph_reshaping_walk(nstack, root(), frc); // Check for unreachable (from below) code (i.e., infinite loops). - for( uint i = 0; i < fpu._tests.size(); i++ ) { - MultiBranchNode *n = fpu._tests[i]->as_MultiBranch(); + for( uint i = 0; i < frc._tests.size(); i++ ) { + MultiBranchNode *n = frc._tests[i]->as_MultiBranch(); // Get number of CFG targets. // Note that PCTables include exception targets after calls. uint required_outcnt = n->required_outcnt(); @@ -2497,7 +2512,7 @@ // Check that I actually visited all kids. Unreached kids // must be infinite loops. for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) - if (!fpu._visited.test(n->fast_out(j)->_idx)) { + if (!frc._visited.test(n->fast_out(j)->_idx)) { record_method_not_compilable("infinite loop"); return true; // Found unvisited kid; must be unreach } @@ -2506,13 +2521,14 @@ // If original bytecodes contained a mixture of floats and doubles // check if the optimizer has made it homogenous, item (3). if( Use24BitFPMode && Use24BitFP && - fpu.get_float_count() > 32 && - fpu.get_double_count() == 0 && - (10 * fpu.get_call_count() < fpu.get_float_count()) ) { + frc.get_float_count() > 32 && + frc.get_double_count() == 0 && + (10 * frc.get_call_count() < frc.get_float_count()) ) { set_24_bit_selection_and_mode( false, true ); } - set_has_java_calls(fpu.get_java_call_count() > 0); + set_java_calls(frc.get_java_call_count()); + set_inner_loops(frc.get_inner_loop_count()); // No infinite loops, no reason to bail out. return false; diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/compile.hpp --- a/src/share/vm/opto/compile.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/compile.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -223,7 +223,8 @@ PhaseCFG* _cfg; // Results of CFG finding bool _select_24_bit_instr; // We selected an instruction with a 24-bit result bool _in_24_bit_fp_mode; // We are emitting instructions with 24-bit results - bool _has_java_calls; // True if the method has java calls + int _java_calls; // Number of java calls in the method + int _inner_loops; // Number of inner loops in the method Matcher* _matcher; // Engine to map ideal to machine instructions PhaseRegAlloc* _regalloc; // Results of register allocation. int _frame_slots; // Size of total frame in stack slots @@ -505,7 +506,9 @@ PhaseCFG* cfg() { return _cfg; } bool select_24_bit_instr() const { return _select_24_bit_instr; } bool in_24_bit_fp_mode() const { return _in_24_bit_fp_mode; } - bool has_java_calls() const { return _has_java_calls; } + bool has_java_calls() const { return _java_calls > 0; } + int java_calls() const { return _java_calls; } + int inner_loops() const { return _inner_loops; } Matcher* matcher() { return _matcher; } PhaseRegAlloc* regalloc() { return _regalloc; } int frame_slots() const { return _frame_slots; } @@ -532,7 +535,8 @@ _in_24_bit_fp_mode = mode; } - void set_has_java_calls(bool z) { _has_java_calls = z; } + void set_java_calls(int z) { _java_calls = z; } + void set_inner_loops(int z) { _inner_loops = z; } // Instruction bits passed off to the VM int code_size() { return _method_size; } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/escape.cpp --- a/src/share/vm/opto/escape.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/escape.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -578,11 +578,24 @@ if (phi_alias_idx == alias_idx) { return orig_phi; } - // have we already created a Phi for this alias index? + // Have we recently created a Phi for this alias index? PhiNode *result = get_map_phi(orig_phi->_idx); if (result != NULL && C->get_alias_index(result->adr_type()) == alias_idx) { return result; } + // Previous check may fail when the same wide memory Phi was split into Phis + // for different memory slices. Search all Phis for this region. + if (result != NULL) { + Node* region = orig_phi->in(0); + for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) { + Node* phi = region->fast_out(i); + if (phi->is_Phi() && + C->get_alias_index(phi->as_Phi()->adr_type()) == alias_idx) { + assert(phi->_idx >= nodes_size(), "only new Phi per instance memory slice"); + return phi->as_Phi(); + } + } + } if ((int)C->unique() + 2*NodeLimitFudgeFactor > MaxNodeLimit) { if (C->do_escape_analysis() == true && !C->failing()) { // Retry compilation without escape analysis. @@ -595,6 +608,7 @@ orig_phi_worklist.append_if_missing(orig_phi); const TypePtr *atype = C->get_adr_type(alias_idx); result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype); + C->copy_node_notes_to(result, orig_phi); set_map_phi(orig_phi->_idx, result); igvn->set_type(result, result->bottom_type()); record_for_optimizer(result); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/graphKit.cpp --- a/src/share/vm/opto/graphKit.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/graphKit.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -1373,11 +1373,12 @@ return st; } + void GraphKit::pre_barrier(Node* ctl, Node* obj, Node* adr, - uint adr_idx, - Node *val, + uint adr_idx, + Node* val, const TypeOopPtr* val_type, BasicType bt) { BarrierSet* bs = Universe::heap()->barrier_set(); @@ -1385,7 +1386,7 @@ switch (bs->kind()) { case BarrierSet::G1SATBCT: case BarrierSet::G1SATBCTLogging: - g1_write_barrier_pre(obj, adr, adr_idx, val, val_type, bt); + g1_write_barrier_pre(obj, adr, adr_idx, val, val_type, bt); break; case BarrierSet::CardTableModRef: @@ -1404,8 +1405,8 @@ Node* store, Node* obj, Node* adr, - uint adr_idx, - Node *val, + uint adr_idx, + Node* val, BasicType bt, bool use_precise) { BarrierSet* bs = Universe::heap()->barrier_set(); @@ -1413,7 +1414,7 @@ switch (bs->kind()) { case BarrierSet::G1SATBCT: case BarrierSet::G1SATBCTLogging: - g1_write_barrier_post(store, obj, adr, adr_idx, val, bt, use_precise); + g1_write_barrier_post(store, obj, adr, adr_idx, val, bt, use_precise); break; case BarrierSet::CardTableModRef: @@ -1431,42 +1432,36 @@ } } -Node* GraphKit::store_oop_to_object(Node* ctl, - Node* obj, - Node* adr, - const TypePtr* adr_type, - Node *val, - const TypeOopPtr* val_type, - BasicType bt) { +Node* GraphKit::store_oop(Node* ctl, + Node* obj, + Node* adr, + const TypePtr* adr_type, + Node* val, + const TypeOopPtr* val_type, + BasicType bt, + bool use_precise) { + + set_control(ctl); + if (stopped()) return top(); // Dead path ? + + assert(bt == T_OBJECT, "sanity"); + assert(val != NULL, "not dead path"); uint adr_idx = C->get_alias_index(adr_type); - Node* store; - pre_barrier(ctl, obj, adr, adr_idx, val, val_type, bt); - store = store_to_memory(control(), adr, val, bt, adr_idx); - post_barrier(control(), store, obj, adr, adr_idx, val, bt, false); + assert(adr_idx != Compile::AliasIdxTop, "use other store_to_memory factory" ); + + pre_barrier(control(), obj, adr, adr_idx, val, val_type, bt); + Node* store = store_to_memory(control(), adr, val, bt, adr_idx); + post_barrier(control(), store, obj, adr, adr_idx, val, bt, use_precise); return store; } -Node* GraphKit::store_oop_to_array(Node* ctl, - Node* obj, - Node* adr, - const TypePtr* adr_type, - Node *val, - const TypeOopPtr* val_type, - BasicType bt) { - uint adr_idx = C->get_alias_index(adr_type); - Node* store; - pre_barrier(ctl, obj, adr, adr_idx, val, val_type, bt); - store = store_to_memory(control(), adr, val, bt, adr_idx); - post_barrier(control(), store, obj, adr, adr_idx, val, bt, true); - return store; -} - +// Could be an array or object we don't know at compile time (unsafe ref.) Node* GraphKit::store_oop_to_unknown(Node* ctl, - Node* obj, - Node* adr, - const TypePtr* adr_type, - Node *val, - BasicType bt) { + Node* obj, // containing obj + Node* adr, // actual adress to store val at + const TypePtr* adr_type, + Node* val, + BasicType bt) { Compile::AliasType* at = C->alias_type(adr_type); const TypeOopPtr* val_type = NULL; if (adr_type->isa_instptr()) { @@ -1485,12 +1480,7 @@ if (val_type == NULL) { val_type = TypeInstPtr::BOTTOM; } - - uint adr_idx = at->index(); - pre_barrier(ctl, obj, adr, adr_idx, val, val_type, bt); - Node* store = store_to_memory(control(), adr, val, bt, adr_idx); - post_barrier(control(), store, obj, adr, adr_idx, val, bt, true); - return store; + return store_oop(ctl, obj, adr, adr_type, val, val_type, bt, true); } @@ -1804,93 +1794,6 @@ } -//------------------------------store_barrier---------------------------------- -// Insert a write-barrier store. This is to let generational GC work; we have -// to flag all oop-stores before the next GC point. -void GraphKit::write_barrier_post(Node* oop_store, Node* obj, Node* adr, - Node* val, bool use_precise) { - // No store check needed if we're storing a NULL or an old object - // (latter case is probably a string constant). The concurrent - // mark sweep garbage collector, however, needs to have all nonNull - // oop updates flagged via card-marks. - if (val != NULL && val->is_Con()) { - // must be either an oop or NULL - const Type* t = val->bottom_type(); - if (t == TypePtr::NULL_PTR || t == Type::TOP) - // stores of null never (?) need barriers - return; - ciObject* con = t->is_oopptr()->const_oop(); - if (con != NULL - && con->is_perm() - && Universe::heap()->can_elide_permanent_oop_store_barriers()) - // no store barrier needed, because no old-to-new ref created - return; - } - - if (use_ReduceInitialCardMarks() - && obj == just_allocated_object(control())) { - // We can skip marks on a freshly-allocated object. - // Keep this code in sync with do_eager_card_mark in runtime.cpp. - // That routine eagerly marks the occasional object which is produced - // by the slow path, so that we don't have to do it here. - return; - } - - if (!use_precise) { - // All card marks for a (non-array) instance are in one place: - adr = obj; - } - // (Else it's an array (or unknown), and we want more precise card marks.) - assert(adr != NULL, ""); - - // Get the alias_index for raw card-mark memory - int adr_type = Compile::AliasIdxRaw; - // Convert the pointer to an int prior to doing math on it - Node* cast = _gvn.transform(new (C, 2) CastP2XNode(control(), adr)); - // Divide by card size - assert(Universe::heap()->barrier_set()->kind() == BarrierSet::CardTableModRef, - "Only one we handle so far."); - CardTableModRefBS* ct = - (CardTableModRefBS*)(Universe::heap()->barrier_set()); - Node *b = _gvn.transform(new (C, 3) URShiftXNode( cast, _gvn.intcon(CardTableModRefBS::card_shift) )); - // We store into a byte array, so do not bother to left-shift by zero - Node *c = byte_map_base_node(); - // Combine - Node *sb_ctl = control(); - Node *sb_adr = _gvn.transform(new (C, 4) AddPNode( top()/*no base ptr*/, c, b )); - Node *sb_val = _gvn.intcon(0); - // Smash zero into card - if( !UseConcMarkSweepGC ) { - BasicType bt = T_BYTE; - store_to_memory(sb_ctl, sb_adr, sb_val, bt, adr_type); - } else { - // Specialized path for CM store barrier - cms_card_mark( sb_ctl, sb_adr, sb_val, oop_store); - } -} - -// Specialized path for CMS store barrier -void GraphKit::cms_card_mark(Node* ctl, Node* adr, Node* val, Node *oop_store) { - BasicType bt = T_BYTE; - int adr_idx = Compile::AliasIdxRaw; - Node* mem = memory(adr_idx); - - // The type input is NULL in PRODUCT builds - const TypePtr* type = NULL; - debug_only(type = C->get_adr_type(adr_idx)); - - // Add required edge to oop_store, optimizer does not support precedence edges. - // Convert required edge to precedence edge before allocation. - Node *store = _gvn.transform( new (C, 5) StoreCMNode(ctl, mem, adr, type, val, oop_store) ); - set_memory(store, adr_idx); - - // For CMS, back-to-back card-marks can only remove the first one - // and this requires DU info. Push on worklist for optimizer. - if (mem->req() > MemNode::Address && adr == mem->in(MemNode::Address)) - record_for_igvn(store); -} - - void GraphKit::round_double_arguments(ciMethod* dest_method) { // (Note: TypeFunc::make has a cache that makes this fast.) const TypeFunc* tf = TypeFunc::make(dest_method); @@ -3215,6 +3118,79 @@ return NULL; } +//----------------------------- store barriers ---------------------------- +#define __ ideal. + +void GraphKit::sync_kit(IdealKit& ideal) { + // Final sync IdealKit and graphKit. + __ drain_delay_transform(); + set_all_memory(__ merged_memory()); + set_control(__ ctrl()); +} + +// vanilla/CMS post barrier +// Insert a write-barrier store. This is to let generational GC work; we have +// to flag all oop-stores before the next GC point. +void GraphKit::write_barrier_post(Node* oop_store, + Node* obj, + Node* adr, + Node* val, + bool use_precise) { + // No store check needed if we're storing a NULL or an old object + // (latter case is probably a string constant). The concurrent + // mark sweep garbage collector, however, needs to have all nonNull + // oop updates flagged via card-marks. + if (val != NULL && val->is_Con()) { + // must be either an oop or NULL + const Type* t = val->bottom_type(); + if (t == TypePtr::NULL_PTR || t == Type::TOP) + // stores of null never (?) need barriers + return; + ciObject* con = t->is_oopptr()->const_oop(); + if (con != NULL + && con->is_perm() + && Universe::heap()->can_elide_permanent_oop_store_barriers()) + // no store barrier needed, because no old-to-new ref created + return; + } + + if (!use_precise) { + // All card marks for a (non-array) instance are in one place: + adr = obj; + } + // (Else it's an array (or unknown), and we want more precise card marks.) + assert(adr != NULL, ""); + + IdealKit ideal(gvn(), control(), merged_memory(), true); + + // Convert the pointer to an int prior to doing math on it + Node* cast = __ CastPX(__ ctrl(), adr); + + // Divide by card size + assert(Universe::heap()->barrier_set()->kind() == BarrierSet::CardTableModRef, + "Only one we handle so far."); + Node* card_offset = __ URShiftX( cast, __ ConI(CardTableModRefBS::card_shift) ); + + // Combine card table base and card offset + Node* card_adr = __ AddP(__ top(), byte_map_base_node(), card_offset ); + + // Get the alias_index for raw card-mark memory + int adr_type = Compile::AliasIdxRaw; + // Smash zero into card + Node* zero = __ ConI(0); + BasicType bt = T_BYTE; + if( !UseConcMarkSweepGC ) { + __ store(__ ctrl(), card_adr, zero, bt, adr_type); + } else { + // Specialized path for CM store barrier + __ storeCM(__ ctrl(), card_adr, zero, oop_store, bt, adr_type); + } + + // Final sync IdealKit and GraphKit. + sync_kit(ideal); +} + +// G1 pre/post barriers void GraphKit::g1_write_barrier_pre(Node* obj, Node* adr, uint alias_idx, @@ -3222,10 +3198,8 @@ const TypeOopPtr* val_type, BasicType bt) { IdealKit ideal(gvn(), control(), merged_memory(), true); -#define __ ideal. - __ declares_done(); - - Node* thread = __ thread(); + + Node* tls = __ thread(); // ThreadLocalStorage Node* no_ctrl = NULL; Node* no_base = __ top(); @@ -3248,9 +3222,9 @@ // set_control( ctl); - Node* marking_adr = __ AddP(no_base, thread, __ ConX(marking_offset)); - Node* buffer_adr = __ AddP(no_base, thread, __ ConX(buffer_offset)); - Node* index_adr = __ AddP(no_base, thread, __ ConX(index_offset)); + Node* marking_adr = __ AddP(no_base, tls, __ ConX(marking_offset)); + Node* buffer_adr = __ AddP(no_base, tls, __ ConX(buffer_offset)); + Node* index_adr = __ AddP(no_base, tls, __ ConX(index_offset)); // Now some of the values @@ -3278,55 +3252,52 @@ Node* next_index = __ SubI(index, __ ConI(sizeof(intptr_t))); Node* next_indexX = next_index; #ifdef _LP64 - // We could refine the type for what it's worth - // const TypeLong* lidxtype = TypeLong::make(CONST64(0), get_size_from_queue); - next_indexX = _gvn.transform( new (C, 2) ConvI2LNode(next_index, TypeLong::make(0, max_jlong, Type::WidenMax)) ); -#endif // _LP64 + // We could refine the type for what it's worth + // const TypeLong* lidxtype = TypeLong::make(CONST64(0), get_size_from_queue); + next_indexX = _gvn.transform( new (C, 2) ConvI2LNode(next_index, TypeLong::make(0, max_jlong, Type::WidenMax)) ); +#endif // Now get the buffer location we will log the original value into and store it - Node *log_addr = __ AddP(no_base, buffer, next_indexX); - // __ store(__ ctrl(), log_addr, orig, T_OBJECT, C->get_alias_index(TypeOopPtr::BOTTOM)); __ store(__ ctrl(), log_addr, orig, T_OBJECT, Compile::AliasIdxRaw); - // update the index - // __ store(__ ctrl(), index_adr, next_index, T_INT, Compile::AliasIdxRaw); - // This is a hack to force this store to occur before the oop store that is coming up - __ store(__ ctrl(), index_adr, next_index, T_INT, C->get_alias_index(TypeOopPtr::BOTTOM)); + __ store(__ ctrl(), index_adr, next_index, T_INT, Compile::AliasIdxRaw); } __ else_(); { // logging buffer is full, call the runtime const TypeFunc *tf = OptoRuntime::g1_wb_pre_Type(); - // __ make_leaf_call(tf, OptoRuntime::g1_wb_pre_Java(), "g1_wb_pre", orig, thread); - __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), "g1_wb_pre", orig, thread); - } __ end_if(); - } __ end_if(); - } __ end_if(); - - __ drain_delay_transform(); - set_control( __ ctrl()); - set_all_memory( __ merged_memory()); - -#undef __ + __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), "g1_wb_pre", orig, tls); + } __ end_if(); // (!index) + } __ end_if(); // (orig != NULL) + } __ end_if(); // (!marking) + + // Final sync IdealKit and GraphKit. + sync_kit(ideal); } // // Update the card table and add card address to the queue // -void GraphKit::g1_mark_card(IdealKit* ideal, Node* card_adr, Node* store, Node* index, Node* index_adr, Node* buffer, const TypeFunc* tf) { -#define __ ideal-> +void GraphKit::g1_mark_card(IdealKit& ideal, + Node* card_adr, + Node* oop_store, + Node* index, + Node* index_adr, + Node* buffer, + const TypeFunc* tf) { + Node* zero = __ ConI(0); Node* no_base = __ top(); BasicType card_bt = T_BYTE; // Smash zero into card. MUST BE ORDERED WRT TO STORE - __ storeCM(__ ctrl(), card_adr, zero, store, card_bt, Compile::AliasIdxRaw); + __ storeCM(__ ctrl(), card_adr, zero, oop_store, card_bt, Compile::AliasIdxRaw); // Now do the queue work __ if_then(index, BoolTest::ne, zero); { - Node* next_index = __ SubI(index, __ ConI(sizeof(intptr_t))); + Node* next_index = __ SubI(index, __ ConI(sizeof(intptr_t))); Node* next_indexX = next_index; #ifdef _LP64 // We could refine the type for what it's worth @@ -3341,10 +3312,10 @@ } __ else_(); { __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), "g1_wb_post", card_adr, __ thread()); } __ end_if(); -#undef __ + } -void GraphKit::g1_write_barrier_post(Node* store, +void GraphKit::g1_write_barrier_post(Node* oop_store, Node* obj, Node* adr, uint alias_idx, @@ -3369,10 +3340,8 @@ assert(adr != NULL, ""); IdealKit ideal(gvn(), control(), merged_memory(), true); -#define __ ideal. - __ declares_done(); - - Node* thread = __ thread(); + + Node* tls = __ thread(); // ThreadLocalStorage Node* no_ctrl = NULL; Node* no_base = __ top(); @@ -3394,8 +3363,8 @@ // Pointers into the thread - Node* buffer_adr = __ AddP(no_base, thread, __ ConX(buffer_offset)); - Node* index_adr = __ AddP(no_base, thread, __ ConX(index_offset)); + Node* buffer_adr = __ AddP(no_base, tls, __ ConX(buffer_offset)); + Node* index_adr = __ AddP(no_base, tls, __ ConX(index_offset)); // Now some values @@ -3404,18 +3373,14 @@ // Convert the store obj pointer to an int prior to doing math on it - // Use addr not obj gets accurate card marks - - // Node* cast = __ CastPX(no_ctrl, adr /* obj */); - // Must use ctrl to prevent "integerized oop" existing across safepoint - Node* cast = __ CastPX(__ ctrl(), ( use_precise ? adr : obj )); + Node* cast = __ CastPX(__ ctrl(), adr); // Divide pointer by card size Node* card_offset = __ URShiftX( cast, __ ConI(CardTableModRefBS::card_shift) ); // Combine card table base and card offset - Node *card_adr = __ AddP(no_base, byte_map_base_node(), card_offset ); + Node* card_adr = __ AddP(no_base, byte_map_base_node(), card_offset ); // If we know the value being stored does it cross regions? @@ -3439,18 +3404,17 @@ Node* card_val = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw); __ if_then(card_val, BoolTest::ne, zero); { - g1_mark_card(&ideal, card_adr, store, index, index_adr, buffer, tf); + g1_mark_card(ideal, card_adr, oop_store, index, index_adr, buffer, tf); } __ end_if(); } __ end_if(); } __ end_if(); } else { - g1_mark_card(&ideal, card_adr, store, index, index_adr, buffer, tf); + // Object.clone() instrinsic uses this path. + g1_mark_card(ideal, card_adr, oop_store, index, index_adr, buffer, tf); } - - __ drain_delay_transform(); - set_control( __ ctrl()); - set_all_memory( __ merged_memory()); + // Final sync IdealKit and GraphKit. + sync_kit(ideal); +} #undef __ -} diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/graphKit.hpp --- a/src/share/vm/opto/graphKit.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/graphKit.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -449,13 +449,24 @@ // // If val==NULL, it is taken to be a completely unknown value. QQQ + Node* store_oop(Node* ctl, + Node* obj, // containing obj + Node* adr, // actual adress to store val at + const TypePtr* adr_type, + Node* val, + const TypeOopPtr* val_type, + BasicType bt, + bool use_precise); + Node* store_oop_to_object(Node* ctl, Node* obj, // containing obj Node* adr, // actual adress to store val at const TypePtr* adr_type, Node* val, const TypeOopPtr* val_type, - BasicType bt); + BasicType bt) { + return store_oop(ctl, obj, adr, adr_type, val, val_type, bt, false); + } Node* store_oop_to_array(Node* ctl, Node* obj, // containing obj @@ -463,7 +474,9 @@ const TypePtr* adr_type, Node* val, const TypeOopPtr* val_type, - BasicType bt); + BasicType bt) { + return store_oop(ctl, obj, adr, adr_type, val, val_type, bt, true); + } // Could be an array or object we don't know at compile time (unsafe ref.) Node* store_oop_to_unknown(Node* ctl, @@ -488,9 +501,6 @@ // Return a load of array element at idx. Node* load_array_element(Node* ctl, Node* ary, Node* idx, const TypeAryPtr* arytype); - // CMS card-marks have an input from the corresponding oop_store - void cms_card_mark(Node* ctl, Node* adr, Node* val, Node* oop_store); - //---------------- Dtrace support -------------------- void make_dtrace_method_entry_exit(ciMethod* method, bool is_entry); void make_dtrace_method_entry(ciMethod* method) { @@ -582,9 +592,6 @@ return C->too_many_recompiles(method(), bci(), reason); } - // vanilla/CMS post barrier - void write_barrier_post(Node *store, Node* obj, Node* adr, Node* val, bool use_precise); - // Returns the object (if any) which was created the moment before. Node* just_allocated_object(Node* current_control); @@ -593,6 +600,11 @@ && Universe::heap()->can_elide_tlab_store_barriers()); } + void sync_kit(IdealKit& ideal); + + // vanilla/CMS post barrier + void write_barrier_post(Node *store, Node* obj, Node* adr, Node* val, bool use_precise); + // G1 pre/post barriers void g1_write_barrier_pre(Node* obj, Node* adr, @@ -610,7 +622,7 @@ bool use_precise); // Helper function for g1 private: - void g1_mark_card(IdealKit* ideal, Node* card_adr, Node* store, Node* index, Node* index_adr, + void g1_mark_card(IdealKit& ideal, Node* card_adr, Node* store, Node* index, Node* index_adr, Node* buffer, const TypeFunc* tf); public: diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/idealKit.cpp --- a/src/share/vm/opto/idealKit.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/idealKit.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -34,7 +34,7 @@ const uint IdealKit::first_var = TypeFunc::Parms + 1; //----------------------------IdealKit----------------------------------------- -IdealKit::IdealKit(PhaseGVN &gvn, Node* control, Node* mem, bool delay_all_transforms) : +IdealKit::IdealKit(PhaseGVN &gvn, Node* control, Node* mem, bool delay_all_transforms, bool has_declarations) : _gvn(gvn), C(gvn.C) { _initial_ctrl = control; _initial_memory = mem; @@ -47,6 +47,9 @@ _pending_cvstates = new (C->node_arena()) GrowableArray(C->node_arena(), init_size, 0, 0); _delay_transform = new (C->node_arena()) GrowableArray(C->node_arena(), init_size, 0, 0); DEBUG_ONLY(_state = new (C->node_arena()) GrowableArray(C->node_arena(), init_size, 0, 0)); + if (!has_declarations) { + declarations_done(); + } } //-------------------------------if_then------------------------------------- @@ -97,7 +100,7 @@ //-------------------------------end_if------------------------------------- // Merge the "then" and "else" cvstates. // -// The if_then() pushed the current state for later use +// The if_then() pushed a copy of the current state for later use // as the initial state for a future "else" clause. The // current state then became the initial state for the // then clause. If an "else" clause was encountered, it will @@ -258,8 +261,8 @@ return delay_transform(PhiNode::make(reg, n, ct)); } -//-----------------------------declares_done----------------------------------- -void IdealKit::declares_done() { +//-----------------------------declarations_done------------------------------- +void IdealKit::declarations_done() { _cvstate = new_cvstate(); // initialize current cvstate set_ctrl(_initial_ctrl); // initialize control in current cvstate set_all_memory(_initial_memory);// initialize memory in current cvstate @@ -277,7 +280,9 @@ //-----------------------------delay_transform----------------------------------- Node* IdealKit::delay_transform(Node* n) { - gvn().set_type(n, n->bottom_type()); + if (!gvn().is_IterGVN() || !gvn().is_IterGVN()->delay_transform()) { + gvn().set_type(n, n->bottom_type()); + } _delay_transform->push(n); return n; } @@ -321,7 +326,9 @@ Node* IdealKit::memory(uint alias_idx) { MergeMemNode* mem = merged_memory(); Node* p = mem->memory_at(alias_idx); - _gvn.set_type(p, Type::MEMORY); // must be mapped + if (!gvn().is_IterGVN() || !gvn().is_IterGVN()->delay_transform()) { + _gvn.set_type(p, Type::MEMORY); // must be mapped + } return p; } @@ -462,9 +469,6 @@ const TypePtr* adr_type = TypeRawPtr::BOTTOM; uint adr_idx = C->get_alias_index(adr_type); - // Clone initial memory - MergeMemNode* cloned_mem = MergeMemNode::make(C, merged_memory()); - // Slow-path leaf call int size = slow_call_type->domain()->cnt(); CallNode *call = (CallNode*)new (C, size) CallLeafNode( slow_call_type, slow_call, leaf_name, adr_type); @@ -489,9 +493,6 @@ set_ctrl(transform( new (C, 1) ProjNode(call,TypeFunc::Control) )); - // Set the incoming clone of memory as current memory - set_all_memory(cloned_mem); - // Make memory for the call Node* mem = _gvn.transform( new (C, 1) ProjNode(call, TypeFunc::Memory) ); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/idealKit.hpp --- a/src/share/vm/opto/idealKit.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/idealKit.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -49,7 +49,7 @@ // Example: // Node* limit = ?? // IdealVariable i(kit), j(kit); -// declares_done(); +// declarations_done(); // Node* exit = make_label(1); // 1 goto // set(j, ConI(0)); // loop(i, ConI(0), BoolTest::lt, limit); { @@ -101,10 +101,7 @@ Node* new_cvstate(); // Create a new cvstate Node* cvstate() { return _cvstate; } // current cvstate Node* copy_cvstate(); // copy current cvstate - void set_ctrl(Node* ctrl) { _cvstate->set_req(TypeFunc::Control, ctrl); } - // Should this assert this is a MergeMem??? - void set_all_memory(Node* mem){ _cvstate->set_req(TypeFunc::Memory, mem); } void set_memory(Node* mem, uint alias_idx ); void do_memory_merge(Node* merging, Node* join); void clear(Node* m); // clear a cvstate @@ -132,15 +129,17 @@ Node* memory(uint alias_idx); public: - IdealKit(PhaseGVN &gvn, Node* control, Node* memory, bool delay_all_transforms = false); + IdealKit(PhaseGVN &gvn, Node* control, Node* memory, bool delay_all_transforms = false, bool has_declarations = false); ~IdealKit() { stop(); drain_delay_transform(); } // Control Node* ctrl() { return _cvstate->in(TypeFunc::Control); } + void set_ctrl(Node* ctrl) { _cvstate->set_req(TypeFunc::Control, ctrl); } Node* top() { return C->top(); } MergeMemNode* merged_memory() { return _cvstate->in(TypeFunc::Memory)->as_MergeMem(); } + void set_all_memory(Node* mem) { _cvstate->set_req(TypeFunc::Memory, mem); } void set(IdealVariable& v, Node* rhs) { _cvstate->set_req(first_var + v.id(), rhs); } Node* value(IdealVariable& v) { return _cvstate->in(first_var + v.id()); } void dead(IdealVariable& v) { set(v, (Node*)NULL); } @@ -155,7 +154,7 @@ Node* make_label(int goto_ct); void bind(Node* lab); void goto_(Node* lab, bool bind = false); - void declares_done(); + void declarations_done(); void drain_delay_transform(); Node* IfTrue(IfNode* iff) { return transform(new (C,1) IfTrueNode(iff)); } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/ifnode.cpp --- a/src/share/vm/opto/ifnode.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/ifnode.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -378,7 +378,18 @@ // Force the original merge dead igvn->hash_delete(r); - r->set_req_X(0,NULL,igvn); + // First, remove region's dead users. + for (DUIterator_Last lmin, l = r->last_outs(lmin); l >= lmin;) { + Node* u = r->last_out(l); + if( u == r ) { + r->set_req(0, NULL); + } else { + assert(u->outcnt() == 0, "only dead users"); + igvn->remove_dead_node(u); + } + l -= 1; + } + igvn->remove_dead_node(r); // Now remove the bogus extra edges used to keep things alive igvn->remove_dead_node( hook ); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/library_call.cpp --- a/src/share/vm/opto/library_call.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/library_call.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -310,11 +310,6 @@ if (!InlineAtomicLong) return NULL; break; - case vmIntrinsics::_Object_init: - case vmIntrinsics::_invoke: - // We do not intrinsify these; they are marked for other purposes. - return NULL; - case vmIntrinsics::_getCallerClass: if (!UseNewReflection) return NULL; if (!InlineReflectionGetCallerClass) return NULL; @@ -327,6 +322,8 @@ break; default: + assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility"); + assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?"); break; } @@ -394,18 +391,11 @@ } if (PrintIntrinsics) { - switch (intrinsic_id()) { - case vmIntrinsics::_invoke: - case vmIntrinsics::_Object_init: - // We do not expect to inline these, so do not produce any noise about them. - break; - default: - tty->print("Did not inline intrinsic %s%s at bci:%d in", - vmIntrinsics::name_at(intrinsic_id()), - (is_virtual() ? " (virtual)" : ""), kit.bci()); - kit.caller()->print_short_name(tty); - tty->print_cr(" (%d bytes)", kit.caller()->code_size()); - } + tty->print("Did not inline intrinsic %s%s at bci:%d in", + vmIntrinsics::name_at(intrinsic_id()), + (is_virtual() ? " (virtual)" : ""), kit.bci()); + kit.caller()->print_short_name(tty); + tty->print_cr(" (%d bytes)", kit.caller()->code_size()); } C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed); return NULL; @@ -1030,7 +1020,7 @@ const TypeAry* target_array_type = TypeAry::make(TypeInt::CHAR, TypeInt::make(0, target_length, Type::WidenMin)); const TypeAryPtr* target_type = TypeAryPtr::make(TypePtr::BotPTR, target_array_type, target_array->klass(), true, Type::OffsetBot); - IdealKit kit(gvn(), control(), merged_memory()); + IdealKit kit(gvn(), control(), merged_memory(), false, true); #define __ kit. Node* zero = __ ConI(0); Node* one = __ ConI(1); @@ -1042,7 +1032,7 @@ Node* targetOffset = __ ConI(targetOffset_i); Node* sourceEnd = __ SubI(__ AddI(sourceOffset, sourceCount), targetCountLess1); - IdealVariable rtn(kit), i(kit), j(kit); __ declares_done(); + IdealVariable rtn(kit), i(kit), j(kit); __ declarations_done(); Node* outer_loop = __ make_label(2 /* goto */); Node* return_ = __ make_label(1); @@ -1079,9 +1069,9 @@ __ bind(outer_loop); }__ end_loop(); __ dead(i); __ bind(return_); - __ drain_delay_transform(); - - set_control(__ ctrl()); + + // Final sync IdealKit and GraphKit. + sync_kit(kit); Node* result = __ value(rtn); #undef __ C->set_has_loops(true); @@ -2183,14 +2173,23 @@ // of it. So we need to emit code to conditionally do the proper type of // store. - IdealKit kit(gvn(), control(), merged_memory()); - kit.declares_done(); + IdealKit ideal(gvn(), control(), merged_memory()); +#define __ ideal. // QQQ who knows what probability is here?? - kit.if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); { - (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type); - } kit.else_(); { - (void) store_to_memory(control(), adr, val, type, adr_type, is_volatile); - } kit.end_if(); + __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); { + // Sync IdealKit and graphKit. + set_all_memory( __ merged_memory()); + set_control(__ ctrl()); + Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type); + // Update IdealKit memory. + __ set_all_memory(merged_memory()); + __ set_ctrl(control()); + } __ else_(); { + __ store(__ ctrl(), adr, val, type, alias_type->index(), is_volatile); + } __ end_if(); + // Final sync IdealKit and GraphKit. + sync_kit(ideal); +#undef __ } } } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/loopopts.cpp --- a/src/share/vm/opto/loopopts.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/loopopts.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -346,7 +346,10 @@ // Yes! Reshape address expression! Node *inv_scale = new (C, 3) LShiftINode( add_invar, scale ); - register_new_node( inv_scale, add_invar_ctrl ); + Node *inv_scale_ctrl = + dom_depth(add_invar_ctrl) > dom_depth(scale_ctrl) ? + add_invar_ctrl : scale_ctrl; + register_new_node( inv_scale, inv_scale_ctrl ); Node *var_scale = new (C, 3) LShiftINode( add_var, scale ); register_new_node( var_scale, n_ctrl ); Node *var_add = new (C, 3) AddINode( var_scale, inv_scale ); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/machnode.cpp --- a/src/share/vm/opto/machnode.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/machnode.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -300,6 +300,12 @@ } } adr_type = t_disp->add_offset(offset); + } else if( base == NULL && offset != 0 && offset != Type::OffsetBot ) { + // Use ideal type if it is oop ptr. + const TypePtr *tp = oper->type()->isa_ptr(); + if( tp != NULL) { + adr_type = tp; + } } } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/macro.cpp --- a/src/share/vm/opto/macro.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/macro.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -198,14 +198,79 @@ } // Eliminate a card mark sequence. p2x is a ConvP2XNode -void PhaseMacroExpand::eliminate_card_mark(Node *p2x) { +void PhaseMacroExpand::eliminate_card_mark(Node* p2x) { assert(p2x->Opcode() == Op_CastP2X, "ConvP2XNode required"); - Node *shift = p2x->unique_out(); - Node *addp = shift->unique_out(); - for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) { - Node *st = addp->last_out(j); - assert(st->is_Store(), "store required"); - _igvn.replace_node(st, st->in(MemNode::Memory)); + if (!UseG1GC) { + // vanilla/CMS post barrier + Node *shift = p2x->unique_out(); + Node *addp = shift->unique_out(); + for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) { + Node *st = addp->last_out(j); + assert(st->is_Store(), "store required"); + _igvn.replace_node(st, st->in(MemNode::Memory)); + } + } else { + // G1 pre/post barriers + assert(p2x->outcnt() == 2, "expects 2 users: Xor and URShift nodes"); + // It could be only one user, URShift node, in Object.clone() instrinsic + // but the new allocation is passed to arraycopy stub and it could not + // be scalar replaced. So we don't check the case. + + // Remove G1 post barrier. + + // Search for CastP2X->Xor->URShift->Cmp path which + // checks if the store done to a different from the value's region. + // And replace Cmp with #0 (false) to collapse G1 post barrier. + Node* xorx = NULL; + for (DUIterator_Fast imax, i = p2x->fast_outs(imax); i < imax; i++) { + Node* u = p2x->fast_out(i); + if (u->Opcode() == Op_XorX) { + xorx = u; + break; + } + } + assert(xorx != NULL, "missing G1 post barrier"); + Node* shift = xorx->unique_out(); + Node* cmpx = shift->unique_out(); + assert(cmpx->is_Cmp() && cmpx->unique_out()->is_Bool() && + cmpx->unique_out()->as_Bool()->_test._test == BoolTest::ne, + "missing region check in G1 post barrier"); + _igvn.replace_node(cmpx, makecon(TypeInt::CC_EQ)); + + // Remove G1 pre barrier. + + // Search "if (marking != 0)" check and set it to "false". + Node* this_region = p2x->in(0); + assert(this_region != NULL, ""); + // There is no G1 pre barrier if previous stored value is NULL + // (for example, after initialization). + if (this_region->is_Region() && this_region->req() == 3) { + int ind = 1; + if (!this_region->in(ind)->is_IfFalse()) { + ind = 2; + } + if (this_region->in(ind)->is_IfFalse()) { + Node* bol = this_region->in(ind)->in(0)->in(1); + assert(bol->is_Bool(), ""); + cmpx = bol->in(1); + if (bol->as_Bool()->_test._test == BoolTest::ne && + cmpx->is_Cmp() && cmpx->in(2) == intcon(0) && + cmpx->in(1)->is_Load()) { + Node* adr = cmpx->in(1)->as_Load()->in(MemNode::Address); + const int marking_offset = in_bytes(JavaThread::satb_mark_queue_offset() + + PtrQueue::byte_offset_of_active()); + if (adr->is_AddP() && adr->in(AddPNode::Base) == top() && + adr->in(AddPNode::Address)->Opcode() == Op_ThreadLocal && + adr->in(AddPNode::Offset) == MakeConX(marking_offset)) { + _igvn.replace_node(cmpx, makecon(TypeInt::CC_EQ)); + } + } + } + } + // Now CastP2X can be removed since it is used only on dead path + // which currently still alive until igvn optimize it. + assert(p2x->unique_out()->Opcode() == Op_URShiftX, ""); + _igvn.replace_node(p2x, top()); } } @@ -760,14 +825,11 @@ if (n->is_Store()) { _igvn.replace_node(n, n->in(MemNode::Memory)); } else { - assert( n->Opcode() == Op_CastP2X, "CastP2X required"); eliminate_card_mark(n); } k -= (oc2 - use->outcnt()); } } else { - assert( !use->is_SafePoint(), "safepoint uses must have been already elimiated"); - assert( use->Opcode() == Op_CastP2X, "CastP2X required"); eliminate_card_mark(use); } j -= (oc1 - res->outcnt()); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/matcher.cpp --- a/src/share/vm/opto/matcher.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/matcher.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -1489,8 +1489,7 @@ #ifdef ASSERT // Verify adr type after matching memory operation const MachOper* oper = mach->memory_operand(); - if (oper != NULL && oper != (MachOper*)-1 && - mach->adr_type() != TypeRawPtr::BOTTOM) { // non-direct addressing mode + if (oper != NULL && oper != (MachOper*)-1) { // It has a unique memory operand. Find corresponding ideal mem node. Node* m = NULL; if (leaf->is_Mem()) { diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/output.cpp --- a/src/share/vm/opto/output.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/output.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -50,6 +50,13 @@ init_scratch_buffer_blob(); if (failing()) return; // Out of memory + // The number of new nodes (mostly MachNop) is proportional to + // the number of java calls and inner loops which are aligned. + if ( C->check_node_count((NodeLimitFudgeFactor + C->java_calls()*3 + + C->inner_loops()*(OptoLoopAlignment-1)), + "out of nodes before code generation" ) ) { + return; + } // Make sure I can find the Start Node Block_Array& bbs = _cfg->_bbs; Block *entry = _cfg->_blocks[1]; @@ -1105,7 +1112,7 @@ uint *call_returns = NEW_RESOURCE_ARRAY(uint, _cfg->_num_blocks+1); uint return_offset = 0; - MachNode *nop = new (this) MachNopNode(); + int nop_size = (new (this) MachNopNode())->size(_regalloc); int previous_offset = 0; int current_offset = 0; @@ -1188,7 +1195,6 @@ } // align the instruction if necessary - int nop_size = nop->size(_regalloc); int padding = mach->compute_padding(current_offset); // Make sure safepoint node for polling is distinct from a call's // return by adding a nop if needed. @@ -1372,7 +1378,6 @@ // If the next block is the top of a loop, pad this block out to align // the loop top a little. Helps prevent pipe stalls at loop back branches. - int nop_size = (new (this) MachNopNode())->size(_regalloc); if( i<_cfg->_num_blocks-1 ) { Block *nb = _cfg->_blocks[i+1]; uint padding = nb->alignment_padding(current_offset); diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/phaseX.hpp --- a/src/share/vm/opto/phaseX.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/phaseX.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -450,6 +450,8 @@ subsume_node(old, nn); } + bool delay_transform() const { return _delay_transform; } + void set_delay_transform(bool delay) { _delay_transform = delay; } diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/opto/type.hpp --- a/src/share/vm/opto/type.hpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/opto/type.hpp Fri Jul 24 09:01:00 2009 -0700 @@ -1216,6 +1216,8 @@ #define Op_AndX Op_AndL #define Op_AddX Op_AddL #define Op_SubX Op_SubL +#define Op_XorX Op_XorL +#define Op_URShiftX Op_URShiftL // conversions #define ConvI2X(x) ConvI2L(x) #define ConvL2X(x) (x) @@ -1258,6 +1260,8 @@ #define Op_AndX Op_AndI #define Op_AddX Op_AddI #define Op_SubX Op_SubI +#define Op_XorX Op_XorI +#define Op_URShiftX Op_URShiftI // conversions #define ConvI2X(x) (x) #define ConvL2X(x) ConvL2I(x) diff -r f0a1cbbaf3c0 -r 433f394ab509 src/share/vm/runtime/stackValue.cpp --- a/src/share/vm/runtime/stackValue.cpp Thu Jul 16 12:38:26 2009 -0700 +++ b/src/share/vm/runtime/stackValue.cpp Fri Jul 24 09:01:00 2009 -0700 @@ -104,7 +104,17 @@ } #endif case Location::oop: { - Handle h(*(oop *)value_addr); // Wrap a handle around the oop + oop val = *(oop *)value_addr; +#ifdef _LP64 + if (Universe::is_narrow_oop_base(val)) { + // Compiled code may produce decoded oop = narrow_oop_base + // when a narrow oop implicit null check is used. + // The narrow_oop_base could be NULL or be the address + // of the page below heap. Use NULL value for both cases. + val = (oop)NULL; + } +#endif + Handle h(val); // Wrap a handle around the oop return new StackValue(h); } case Location::addr: { diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6826736/Test.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6826736/Test.java Fri Jul 24 09:01:00 2009 -0700 @@ -0,0 +1,75 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6826736 + * @summary CMS: core dump with -XX:+UseCompressedOops + * + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -Xbatch -XX:+ScavengeALot -XX:+UseCompressedOops -XX:HeapBaseMinAddress=32g -XX:CompileThreshold=100 -XX:CompileOnly=Test.test -XX:-BlockLayoutRotateLoops -XX:LoopUnrollLimit=0 Test + */ + +public class Test { + int[] arr; + int[] arr2; + int test(int r) { + for (int i = 0; i < 100; i++) { + for (int j = i; j < 100; j++) { + int a = 0; + for (long k = 0; k < 100; k++) { + a += k; + } + if (arr != null) + a = arr[j]; + r += a; + } + } + return r; + } + + public static void main(String[] args) { + int r = 0; + Test t = new Test(); + for (int i = 0; i < 100; i++) { + t.arr = new int[100]; + r = t.test(r); + } + System.out.println("Warmup 1 is done."); + for (int i = 0; i < 100; i++) { + t.arr = null; + r = t.test(r); + } + System.out.println("Warmup 2 is done."); + for (int i = 0; i < 100; i++) { + t.arr = new int[100]; + r = t.test(r); + } + System.out.println("Warmup is done."); + for (int i = 0; i < 100; i++) { + t.arr = new int[1000000]; + t.arr = null; + r = t.test(r); + } + } +} diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6837094/Test.java --- a/test/compiler/6837094/Test.java Thu Jul 16 12:38:26 2009 -0700 +++ b/test/compiler/6837094/Test.java Fri Jul 24 09:01:00 2009 -0700 @@ -1,5 +1,5 @@ /* - * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 2009 Google Inc. All Rights Reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6851282/Test.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6851282/Test.java Fri Jul 24 09:01:00 2009 -0700 @@ -0,0 +1,124 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6851282 + * @summary JIT miscompilation results in null entry in array when using CompressedOops + * + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:+UseCompressedOops Test + */ + +import java.util.ArrayList; +import java.util.List; + +public class Test { + void foo(A a, A[] as) { + for (A a1 : as) { + B[] filtered = a.c(a1); + for (B b : filtered) { + if (b == null) { + System.out.println("bug: b == null"); + System.exit(97); + } + } + } + } + + public static void main(String[] args) { + List as = new ArrayList(); + for (int i = 0; i < 5000; i++) { + List bs = new ArrayList(); + for (int j = i; j < i + 1000; j++) + bs.add(new B(j)); + as.add(new A(bs.toArray(new B[0]))); + } + new Test().foo(as.get(0), as.subList(1, as.size()).toArray(new A[0])); + } +} + +class A { + final B[] bs; + + public A(B[] bs) { + this.bs = bs; + } + + final B[] c(final A a) { + return new BoxedArray(bs).filter(new Function() { + public Boolean apply(B arg) { + for (B b : a.bs) { + if (b.d == arg.d) + return true; + } + return false; + } + }); + } +} + +class BoxedArray { + + private final T[] array; + + BoxedArray(T[] array) { + this.array = array; + } + + public T[] filter(Function function) { + boolean[] include = new boolean[array.length]; + int len = 0; + int i = 0; + while (i < array.length) { + if (function.apply(array[i])) { + include[i] = true; + len += 1; + } + i += 1; + } + T[] result = (T[]) java.lang.reflect.Array.newInstance(array.getClass().getComponentType(), len); + len = 0; + i = 0; + while (len < result.length) { + if (include[i]) { + result[len] = array[i]; + len += 1; + } + i += 1; + } + return result; + } +} + +interface Function { + R apply(T arg); +} + +class B { + final int d; + public B(int d) { + this.d = d; + } +} + diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6857159/Test6857159.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6857159/Test6857159.java Fri Jul 24 09:01:00 2009 -0700 @@ -0,0 +1,68 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6857159 + * @summary local schedule failed with checkcast of Thread.currentThread() + * + * @run shell Test6857159.sh + */ + +public class Test6857159 extends Thread { + static class ct0 extends Test6857159 { + public void message() { + // System.out.println("message"); + } + + public void run() { + message(); + ct0 ct = (ct0) Thread.currentThread(); + ct.message(); + } + } + static class ct1 extends ct0 { + public void message() { + // System.out.println("message"); + } + } + static class ct2 extends ct0 { + public void message() { + // System.out.println("message"); + } + } + + public static void main(String[] args) throws Exception { + for (int i = 0; i < 100000; i++) { + Thread t = null; + switch (i % 3) { + case 0: t = new ct0(); break; + case 1: t = new ct1(); break; + case 2: t = new ct2(); break; + } + t.start(); + t.join(); + } + } +} diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6857159/Test6857159.sh --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6857159/Test6857159.sh Fri Jul 24 09:01:00 2009 -0700 @@ -0,0 +1,65 @@ +#!/bin/sh +# +# Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. +# DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. +# +# This code is free software; you can redistribute it and/or modify it +# under the terms of the GNU General Public License version 2 only, as +# published by the Free Software Foundation. +# +# This code is distributed in the hope that it will be useful, but WITHOUT +# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or +# FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +# version 2 for more details (a copy is included in the LICENSE file that +# accompanied this code). +# +# You should have received a copy of the GNU General Public License version +# 2 along with this work; if not, write to the Free Software Foundation, +# Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. +# +# Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, +# CA 95054 USA or visit www.sun.com if you need additional information or +# have any questions. +# +# + +if [ "${TESTSRC}" = "" ] +then + echo "TESTSRC not set. Test cannot execute. Failed." + exit 1 +fi +echo "TESTSRC=${TESTSRC}" +if [ "${TESTJAVA}" = "" ] +then + echo "TESTJAVA not set. Test cannot execute. Failed." + exit 1 +fi +echo "TESTJAVA=${TESTJAVA}" +if [ "${TESTCLASSES}" = "" ] +then + echo "TESTCLASSES not set. Test cannot execute. Failed." + exit 1 +fi +echo "TESTCLASSES=${TESTCLASSES}" +echo "CLASSPATH=${CLASSPATH}" + +set -x + +cp ${TESTSRC}/Test6857159.java . +cp ${TESTSRC}/Test6857159.sh . + +${TESTJAVA}/bin/javac -d . Test6857159.java + +${TESTJAVA}/bin/java ${TESTVMOPTS} -Xbatch -XX:+PrintCompilation -XX:CompileOnly=Test6857159\$ct.run Test6857159 > test.out 2>&1 + +grep "COMPILE SKIPPED" test.out + +result=$? +if [ $result -eq 1 ] +then + echo "Passed" + exit 0 +else + echo "Failed" + exit 1 +fi diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6859338/Test6859338.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6859338/Test6859338.java Fri Jul 24 09:01:00 2009 -0700 @@ -0,0 +1,51 @@ +/* + * Copyright 2009 Sun Microsystems, Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6859338 + * @summary Assertion failure in sharedRuntime.cpp + * + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -XX:-InlineObjectHash -Xbatch -XX:-ProfileInterpreter Test6859338 + */ + +public class Test6859338 { + static Object[] o = new Object[] { new Object(), null }; + public static void main(String[] args) { + int total = 0; + try { + // Exercise the implicit null check in the unverified entry point + for (int i = 0; i < 40000; i++) { + int limit = o.length; + if (i < 20000) limit = 1; + for (int j = 0; j < limit; j++) { + total += o[j].hashCode(); + } + } + + } catch (NullPointerException e) { + // this is expected. A true failure causes a crash + } + } +} diff -r f0a1cbbaf3c0 -r 433f394ab509 test/compiler/6860469/Test.java --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6860469/Test.java Fri Jul 24 09:01:00 2009 -0700 @@ -0,0 +1,71 @@ +/* + * Copyright 2009 Google Inc. All Rights Reserved. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. + * + * This code is free software; you can redistribute it and/or modify it + * under the terms of the GNU General Public License version 2 only, as + * published by the Free Software Foundation. + * + * This code is distributed in the hope that it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + * version 2 for more details (a copy is included in the LICENSE file that + * accompanied this code). + * + * You should have received a copy of the GNU General Public License version + * 2 along with this work; if not, write to the Free Software Foundation, + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. + * + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. + * + */ + +/** + * @test + * @bug 6860469 + * @summary remix_address_expressions reshapes address expression with bad control + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test.C Test + */ + +public class Test { + + private static final int H = 16; + private static final int F = 9; + + static int[] fl = new int[1 << F]; + + static int C(int ll, int f) { + int max = -1; + int min = H + 1; + + if (ll != 0) { + if (ll < min) { + min = ll; + } + if (ll > max) { + max = ll; + } + } + + if (f > max) { + f = max; + } + if (min > f) { + min = f; + } + + for (int mc = 1 >> max - f; mc <= 0; mc++) { + int i = mc << (32 - f); + fl[i] = max; + } + + return min; + } + + public static void main(String argv[]) { + C(0, 10); + } +}