Mercurial > hg > graal-jvmci-8
changeset 573:8b22ccb5aba2 jdk7-b49
Merge
author | trims |
---|---|
date | Wed, 25 Feb 2009 23:16:22 -0800 |
parents | 1605bb4eb5a7 (diff) d61c7c22b25c (current diff) |
children | dae503d9f04c |
files | |
diffstat | 85 files changed, 1579 insertions(+), 444 deletions(-) [+] |
line wrap: on
line diff
--- a/make/hotspot_version Thu Feb 19 14:08:00 2009 -0800 +++ b/make/hotspot_version Wed Feb 25 23:16:22 2009 -0800 @@ -35,7 +35,7 @@ HS_MAJOR_VER=15 HS_MINOR_VER=0 -HS_BUILD_NUMBER=01 +HS_BUILD_NUMBER=02 JDK_MAJOR_VER=1 JDK_MINOR_VER=7
--- a/src/cpu/sparc/vm/sparc.ad Thu Feb 19 14:08:00 2009 -0800 +++ b/src/cpu/sparc/vm/sparc.ad Wed Feb 25 23:16:22 2009 -0800 @@ -762,7 +762,7 @@ case Assembler::stdf_op3: st_op = Op_StoreD; break; case Assembler::ldsb_op3: ld_op = Op_LoadB; break; - case Assembler::lduh_op3: ld_op = Op_LoadC; break; + case Assembler::lduh_op3: ld_op = Op_LoadUS; break; case Assembler::ldsh_op3: ld_op = Op_LoadS; break; case Assembler::ldx_op3: // may become LoadP or stay LoadI case Assembler::ldsw_op3: // may become LoadP or stay LoadI @@ -3869,6 +3869,8 @@ constraint(ALLOC_IN_RC(dflt_reg)); match(RegD); + match(regD_low); + format %{ %} interface(REG_INTER); %} @@ -3883,7 +3885,7 @@ operand regD_low() %{ constraint(ALLOC_IN_RC(dflt_low_reg)); - match(RegD); + match(regD); format %{ %} interface(REG_INTER); @@ -5314,9 +5316,9 @@ ins_pipe(iload_mask_mem); %} -// Load Char (16bit UNsigned) into a Long Register -instruct loadUCL(iRegL dst, memory mem, immL_FFFF bytemask) %{ - match(Set dst (AndL (ConvI2L (LoadC mem)) bytemask)); +// Load Unsigned Short/Char (16bit UNsigned) into a Long Register +instruct loadUS2L(iRegL dst, memory mem, immL_FFFF bytemask) %{ + match(Set dst (AndL (ConvI2L (LoadUS mem)) bytemask)); ins_cost(MEMORY_REF_COST); size(4); @@ -5326,9 +5328,9 @@ ins_pipe(iload_mask_mem); %} -// Load Char (16bit unsigned) -instruct loadC(iRegI dst, memory mem) %{ - match(Set dst (LoadC mem)); +// Load Unsigned Short/Char (16bit unsigned) +instruct loadUS(iRegI dst, memory mem) %{ + match(Set dst (LoadUS mem)); ins_cost(MEMORY_REF_COST); size(4);
--- a/src/cpu/x86/vm/x86_32.ad Thu Feb 19 14:08:00 2009 -0800 +++ b/src/cpu/x86/vm/x86_32.ad Wed Feb 25 23:16:22 2009 -0800 @@ -6413,9 +6413,9 @@ ins_pipe( ialu_reg_mem ); %} -// Load Char (16bit unsigned) -instruct loadC(eRegI dst, memory mem) %{ - match(Set dst (LoadC mem)); +// Load Unsigned Short/Char (16bit unsigned) +instruct loadUS(eRegI dst, memory mem) %{ + match(Set dst (LoadUS mem)); ins_cost(125); format %{ "MOVZX $dst,$mem" %}
--- a/src/cpu/x86/vm/x86_64.ad Thu Feb 19 14:08:00 2009 -0800 +++ b/src/cpu/x86/vm/x86_64.ad Wed Feb 25 23:16:22 2009 -0800 @@ -6096,25 +6096,25 @@ // ins_pipe(ialu_reg_mem); // %} -// Load Char (16 bit UNsigned) -instruct loadC(rRegI dst, memory mem) -%{ - match(Set dst (LoadC mem)); +// Load Unsigned Short/Char (16 bit UNsigned) +instruct loadUS(rRegI dst, memory mem) +%{ + match(Set dst (LoadUS mem)); ins_cost(125); - format %{ "movzwl $dst, $mem\t# char" %} + format %{ "movzwl $dst, $mem\t# ushort/char" %} opcode(0x0F, 0xB7); ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem)); ins_pipe(ialu_reg_mem); %} -// Load Char (16 bit UNsigned) into long -// instruct loadC2L(rRegL dst, memory mem) +// Load Unsigned Short/Char (16 bit UNsigned) into long +// instruct loadUS2L(rRegL dst, memory mem) // %{ -// match(Set dst (ConvI2L (LoadC mem))); +// match(Set dst (ConvI2L (LoadUS mem))); // ins_cost(125); -// format %{ "movzwl $dst, $mem\t# char -> long" %} +// format %{ "movzwl $dst, $mem\t# ushort/char -> long" %} // opcode(0x0F, 0xB7); // ins_encode(REX_reg_mem(dst, mem), OpcP, OpcS, reg_mem(dst, mem)); // ins_pipe(ialu_reg_mem); @@ -9490,14 +9490,14 @@ %{ match(Set dst (AndL dst src)); - format %{ "movzbq $dst, $src\t# long & 0xFF" %} + format %{ "movzbq $dst, $dst\t# long & 0xFF" %} opcode(0x0F, 0xB6); ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst)); ins_pipe(ialu_reg); %} // And Register with Immediate 65535 -instruct andL_rReg_imm65535(rRegI dst, immL_65535 src) +instruct andL_rReg_imm65535(rRegL dst, immL_65535 src) %{ match(Set dst (AndL dst src));
--- a/src/os/linux/vm/os_linux.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/os/linux/vm/os_linux.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -1432,6 +1432,10 @@ return buf; } +struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { + return localtime_r(clock, res); +} + //////////////////////////////////////////////////////////////////////////////// // runtime exit support
--- a/src/os/solaris/vm/os_solaris.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/os/solaris/vm/os_solaris.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -323,6 +323,10 @@ return (size_t)(base - bottom); } +struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { + return localtime_r(clock, res); +} + // interruptible infrastructure // setup_interruptible saves the thread state before going into an
--- a/src/os/windows/vm/os_windows.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/os/windows/vm/os_windows.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -327,6 +327,14 @@ return sz; } +struct tm* os::localtime_pd(const time_t* clock, struct tm* res) { + const struct tm* time_struct_ptr = localtime(clock); + if (time_struct_ptr != NULL) { + *res = *time_struct_ptr; + return res; + } + return NULL; +} LONG WINAPI topLevelExceptionFilter(struct _EXCEPTION_POINTERS* exceptionInfo);
--- a/src/share/vm/adlc/dict2.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/adlc/dict2.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -316,9 +316,12 @@ return strcmp((const char *)k1,(const char *)k2); } -// Slimey cheap key comparator. +// Cheap key comparator. int cmpkey(const void *key1, const void *key2) { - return (int)((intptr_t)key1 - (intptr_t)key2); + if (key1 == key2) return 0; + intptr_t delta = (intptr_t)key1 - (intptr_t)key2; + if (delta > 0) return 1; + return -1; } //=============================================================================
--- a/src/share/vm/adlc/forms.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/adlc/forms.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -248,7 +248,7 @@ // True if 'opType', an ideal name, loads or stores. Form::DataType Form::is_load_from_memory(const char *opType) const { if( strcmp(opType,"LoadB")==0 ) return Form::idealB; - if( strcmp(opType,"LoadC")==0 ) return Form::idealC; + if( strcmp(opType,"LoadUS")==0 ) return Form::idealC; if( strcmp(opType,"LoadD")==0 ) return Form::idealD; if( strcmp(opType,"LoadD_unaligned")==0 ) return Form::idealD; if( strcmp(opType,"LoadF")==0 ) return Form::idealF;
--- a/src/share/vm/adlc/formssel.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/adlc/formssel.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -3314,7 +3314,7 @@ "StoreI","StoreL","StoreP","StoreN","StoreD","StoreF" , "StoreB","StoreC","Store" ,"StoreFP", "LoadI" ,"LoadL", "LoadP" ,"LoadN", "LoadD" ,"LoadF" , - "LoadB" ,"LoadC" ,"LoadS" ,"Load" , + "LoadB" ,"LoadUS" ,"LoadS" ,"Load" , "Store4I","Store2I","Store2L","Store2D","Store4F","Store2F","Store16B", "Store8B","Store4B","Store8C","Store4C","Store2C", "Load4I" ,"Load2I" ,"Load2L" ,"Load2D" ,"Load4F" ,"Load2F" ,"Load16B" ,
--- a/src/share/vm/asm/codeBuffer.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/asm/codeBuffer.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -123,6 +123,10 @@ // addresses constructed before expansions will not be confused. cb->free_blob(); } + + // free any overflow storage + delete _overflow_arena; + #ifdef ASSERT Copy::fill_to_bytes(this, sizeof(*this), badResourceValue); #endif
--- a/src/share/vm/classfile/systemDictionary.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/classfile/systemDictionary.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -846,6 +846,12 @@ Handle protection_domain, TRAPS) { + // UseNewReflection + // The result of this call should be consistent with the result + // of the call to resolve_instance_class_or_null(). + // See evaluation 6790209 and 4474172 for more details. + class_loader = Handle(THREAD, java_lang_ClassLoader::non_reflection_class_loader(class_loader())); + unsigned int d_hash = dictionary()->compute_hash(class_name, class_loader); int d_index = dictionary()->hash_to_index(d_hash);
--- a/src/share/vm/gc_implementation/g1/collectionSetChooser.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/collectionSetChooser.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -24,7 +24,7 @@ // We need to sort heap regions by collection desirability. -class CSetChooserCache { +class CSetChooserCache VALUE_OBJ_CLASS_SPEC { private: enum { CacheLength = 16
--- a/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/concurrentG1Refine.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -33,7 +33,7 @@ PYA_cancel // It's been completed by somebody else: cancel. }; -class ConcurrentG1Refine { +class ConcurrentG1Refine: public CHeapObj { ConcurrentG1RefineThread* _cg1rThread; volatile jint _pya;
--- a/src/share/vm/gc_implementation/g1/concurrentMark.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/concurrentMark.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -30,7 +30,7 @@ // A generic CM bit map. This is essentially a wrapper around the BitMap // class, with one bit per (1<<_shifter) HeapWords. -class CMBitMapRO { +class CMBitMapRO VALUE_OBJ_CLASS_SPEC { protected: HeapWord* _bmStartWord; // base address of range covered by map size_t _bmWordSize; // map size (in #HeapWords covered) @@ -139,7 +139,7 @@ // Represents a marking stack used by the CM collector. // Ideally this should be GrowableArray<> just like MSC's marking stack(s). -class CMMarkStack { +class CMMarkStack VALUE_OBJ_CLASS_SPEC { ConcurrentMark* _cm; oop* _base; // bottom of stack jint _index; // one more than last occupied index @@ -237,7 +237,7 @@ void oops_do(OopClosure* f); }; -class CMRegionStack { +class CMRegionStack VALUE_OBJ_CLASS_SPEC { MemRegion* _base; jint _capacity; jint _index; @@ -312,7 +312,7 @@ class ConcurrentMarkThread; -class ConcurrentMark { +class ConcurrentMark: public CHeapObj { friend class ConcurrentMarkThread; friend class CMTask; friend class CMBitMapClosure;
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -141,7 +141,7 @@ _scan_only_head(NULL), _scan_only_tail(NULL), _curr_scan_only(NULL), _length(0), _scan_only_length(0), _last_sampled_rs_lengths(0), - _survivor_head(NULL), _survivors_tail(NULL), _survivor_length(0) + _survivor_head(NULL), _survivor_tail(NULL), _survivor_length(0) { guarantee( check_list_empty(false), "just making sure..." ); } @@ -159,16 +159,15 @@ } void YoungList::add_survivor_region(HeapRegion* hr) { - assert(!hr->is_survivor(), "should not already be for survived"); + assert(hr->is_survivor(), "should be flagged as survivor region"); assert(hr->get_next_young_region() == NULL, "cause it should!"); hr->set_next_young_region(_survivor_head); if (_survivor_head == NULL) { - _survivors_tail = hr; + _survivor_tail = hr; } _survivor_head = hr; - hr->set_survivor(); ++_survivor_length; } @@ -239,7 +238,7 @@ empty_list(_survivor_head); _survivor_head = NULL; - _survivors_tail = NULL; + _survivor_tail = NULL; _survivor_length = 0; _last_sampled_rs_lengths = 0; @@ -391,6 +390,7 @@ // Add survivor regions to SurvRateGroup. _g1h->g1_policy()->note_start_adding_survivor_regions(); + _g1h->g1_policy()->finished_recalculating_age_indexes(true /* is_survivors */); for (HeapRegion* curr = _survivor_head; curr != NULL; curr = curr->get_next_young_region()) { @@ -401,7 +401,7 @@ if (_survivor_head != NULL) { _head = _survivor_head; _length = _survivor_length + _scan_only_length; - _survivors_tail->set_next_young_region(_scan_only_head); + _survivor_tail->set_next_young_region(_scan_only_head); } else { _head = _scan_only_head; _length = _scan_only_length; @@ -418,9 +418,9 @@ _curr_scan_only = NULL; _survivor_head = NULL; - _survivors_tail = NULL; + _survivor_tail = NULL; _survivor_length = 0; - _g1h->g1_policy()->finished_recalculating_age_indexes(); + _g1h->g1_policy()->finished_recalculating_age_indexes(false /* is_survivors */); assert(check_list_well_formed(), "young list should be well formed"); } @@ -553,7 +553,7 @@ if (_gc_alloc_region_counts[purpose] < g1_policy()->max_regions(purpose)) { alloc_region = newAllocRegion_work(word_size, true, zero_filled); if (purpose == GCAllocForSurvived && alloc_region != NULL) { - _young_list->add_survivor_region(alloc_region); + alloc_region->set_survivor(); } ++_gc_alloc_region_counts[purpose]; } else { @@ -949,6 +949,10 @@ GCOverheadReporter::recordSTWEnd(end); g1_policy()->record_full_collection_end(); +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif + gc_epilogue(true); // Abandon concurrent refinement. This must happen last: in the @@ -2593,6 +2597,9 @@ _young_list->print(); #endif // SCAN_ONLY_VERBOSE + g1_policy()->record_survivor_regions(_young_list->survivor_length(), + _young_list->first_survivor_region(), + _young_list->last_survivor_region()); _young_list->reset_auxilary_lists(); } } else { @@ -2619,7 +2626,9 @@ #endif // SCAN_ONLY_VERBOSE double end_time_sec = os::elapsedTime(); - g1_policy()->record_pause_time((end_time_sec - start_time_sec)*1000.0); + if (!evacuation_failed()) { + g1_policy()->record_pause_time((end_time_sec - start_time_sec)*1000.0); + } GCOverheadReporter::recordSTWEnd(end_time_sec); g1_policy()->record_collection_pause_end(popular_region != NULL, abandoned); @@ -2642,8 +2651,13 @@ } } - if (mark_in_progress()) + if (mark_in_progress()) { concurrent_mark()->update_g1_committed(); + } + +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif gc_epilogue(false); } @@ -2754,6 +2768,13 @@ _gc_alloc_region_list = r->next_gc_alloc_region(); r->set_next_gc_alloc_region(NULL); r->set_is_gc_alloc_region(false); + if (r->is_survivor()) { + if (r->is_empty()) { + r->set_not_young(); + } else { + _young_list->add_survivor_region(r); + } + } if (r->is_empty()) { ++_free_regions; } @@ -3150,6 +3171,20 @@ return block; } +void G1CollectedHeap::retire_alloc_region(HeapRegion* alloc_region, + bool par) { + // Another thread might have obtained alloc_region for the given + // purpose, and might be attempting to allocate in it, and might + // succeed. Therefore, we can't do the "finalization" stuff on the + // region below until we're sure the last allocation has happened. + // We ensure this by allocating the remaining space with a garbage + // object. + if (par) par_allocate_remaining_space(alloc_region); + // Now we can do the post-GC stuff on the region. + alloc_region->note_end_of_copying(); + g1_policy()->record_after_bytes(alloc_region->used()); +} + HeapWord* G1CollectedHeap::allocate_during_gc_slow(GCAllocPurpose purpose, HeapRegion* alloc_region, @@ -3167,16 +3202,7 @@ // Otherwise, continue; this new region is empty, too. } assert(alloc_region != NULL, "We better have an allocation region"); - // Another thread might have obtained alloc_region for the given - // purpose, and might be attempting to allocate in it, and might - // succeed. Therefore, we can't do the "finalization" stuff on the - // region below until we're sure the last allocation has happened. - // We ensure this by allocating the remaining space with a garbage - // object. - if (par) par_allocate_remaining_space(alloc_region); - // Now we can do the post-GC stuff on the region. - alloc_region->note_end_of_copying(); - g1_policy()->record_after_bytes(alloc_region->used()); + retire_alloc_region(alloc_region, par); if (_gc_alloc_region_counts[purpose] >= g1_policy()->max_regions(purpose)) { // Cannot allocate more regions for the given purpose. @@ -3185,7 +3211,7 @@ if (purpose != alt_purpose) { HeapRegion* alt_region = _gc_alloc_regions[alt_purpose]; // Has not the alternative region been aliased? - if (alloc_region != alt_region) { + if (alloc_region != alt_region && alt_region != NULL) { // Try to allocate in the alternative region. if (par) { block = alt_region->par_allocate(word_size); @@ -3194,9 +3220,10 @@ } // Make an alias. _gc_alloc_regions[purpose] = _gc_alloc_regions[alt_purpose]; - } - if (block != NULL) { - return block; + if (block != NULL) { + return block; + } + retire_alloc_region(alt_region, par); } // Both the allocation region and the alternative one are full // and aliased, replace them with a new allocation region. @@ -3497,6 +3524,7 @@ OverflowQueue* _overflowed_refs; G1ParGCAllocBuffer _alloc_buffers[GCAllocPurposeCount]; + ageTable _age_table; size_t _alloc_buffer_waste; size_t _undo_waste; @@ -3538,6 +3566,7 @@ _refs(g1h->task_queue(queue_num)), _hash_seed(17), _queue_num(queue_num), _term_attempts(0), + _age_table(false), #if G1_DETAILED_STATS _pushes(0), _pops(0), _steals(0), _steal_attempts(0), _overflow_pushes(0), @@ -3572,8 +3601,9 @@ RefToScanQueue* refs() { return _refs; } OverflowQueue* overflowed_refs() { return _overflowed_refs; } - - inline G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) { + ageTable* age_table() { return &_age_table; } + + G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) { return &_alloc_buffers[purpose]; } @@ -3834,7 +3864,9 @@ (!from_region->is_young() && young_index == 0), "invariant" ); G1CollectorPolicy* g1p = _g1->g1_policy(); markOop m = old->mark(); - GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, m->age(), + int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age() + : m->age(); + GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age, word_sz); HeapWord* obj_ptr = _par_scan_state->allocate(alloc_purpose, word_sz); oop obj = oop(obj_ptr); @@ -3872,9 +3904,12 @@ obj->incr_age(); } else { m = m->incr_age(); + obj->set_mark(m); } + _par_scan_state->age_table()->add(obj, word_sz); + } else { + obj->set_mark(m); } - obj->set_mark(m); // preserve "next" mark bit if (_g1->mark_in_progress() && !_g1->is_obj_ill(old)) { @@ -4129,6 +4164,9 @@ _g1h->g1_policy()->record_obj_copy_time(i, elapsed_ms-term_ms); _g1h->g1_policy()->record_termination_time(i, term_ms); } + if (G1UseSurvivorSpace) { + _g1h->g1_policy()->record_thread_age_table(pss.age_table()); + } _g1h->update_surviving_young_words(pss.surviving_young_words()+1); // Clean up any par-expanded rem sets. @@ -4368,7 +4406,7 @@ // Is this the right thing to do here? We don't save marks // on individual heap regions when we allocate from // them in parallel, so this seems like the correct place for this. - all_alloc_regions_note_end_of_copying(); + retire_all_alloc_regions(); { G1IsAliveClosure is_alive(this); G1KeepAliveClosure keep_alive(this); @@ -5008,7 +5046,7 @@ return no_allocs; } -void G1CollectedHeap::all_alloc_regions_note_end_of_copying() { +void G1CollectedHeap::retire_all_alloc_regions() { for (int ap = 0; ap < GCAllocPurposeCount; ++ap) { HeapRegion* r = _gc_alloc_regions[ap]; if (r != NULL) { @@ -5021,8 +5059,7 @@ } } if (!has_processed_alias) { - r->note_end_of_copying(); - g1_policy()->record_after_bytes(r->used()); + retire_alloc_region(r, false /* par */); } } }
--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -90,7 +90,7 @@ HeapRegion* _curr_scan_only; HeapRegion* _survivor_head; - HeapRegion* _survivors_tail; + HeapRegion* _survivor_tail; size_t _survivor_length; void empty_list(HeapRegion* list); @@ -105,6 +105,7 @@ bool is_empty() { return _length == 0; } size_t length() { return _length; } size_t scan_only_length() { return _scan_only_length; } + size_t survivor_length() { return _survivor_length; } void rs_length_sampling_init(); bool rs_length_sampling_more(); @@ -120,6 +121,7 @@ HeapRegion* first_region() { return _head; } HeapRegion* first_scan_only_region() { return _scan_only_head; } HeapRegion* first_survivor_region() { return _survivor_head; } + HeapRegion* last_survivor_region() { return _survivor_tail; } HeapRegion* par_get_next_scan_only_region() { MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag); HeapRegion* ret = _curr_scan_only; @@ -219,7 +221,7 @@ // The to-space memory regions into which objects are being copied during // a GC. HeapRegion* _gc_alloc_regions[GCAllocPurposeCount]; - uint _gc_alloc_region_counts[GCAllocPurposeCount]; + size_t _gc_alloc_region_counts[GCAllocPurposeCount]; // A list of the regions that have been set to be alloc regions in the // current collection. @@ -281,8 +283,8 @@ // Returns "true" iff none of the gc alloc regions have any allocations // since the last call to "save_marks". bool all_alloc_regions_no_allocs_since_save_marks(); - // Calls "note_end_of_copying on all gc alloc_regions. - void all_alloc_regions_note_end_of_copying(); + // Perform finalization stuff on all allocation regions. + void retire_all_alloc_regions(); // The number of regions allocated to hold humongous objects. int _num_humongous_regions; @@ -351,6 +353,10 @@ // that parallel threads might be attempting allocations. void par_allocate_remaining_space(HeapRegion* r); + // Retires an allocation region when it is full or at the end of a + // GC pause. + void retire_alloc_region(HeapRegion* alloc_region, bool par); + // Helper function for two callbacks below. // "full", if true, indicates that the GC is for a System.gc() request, // and should collect the entire heap. If "clear_all_soft_refs" is true,
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -196,8 +196,13 @@ _short_lived_surv_rate_group(new SurvRateGroup(this, "Short Lived", G1YoungSurvRateNumRegionsSummary)), _survivor_surv_rate_group(new SurvRateGroup(this, "Survivor", - G1YoungSurvRateNumRegionsSummary)) + G1YoungSurvRateNumRegionsSummary)), // add here any more surv rate groups + _recorded_survivor_regions(0), + _recorded_survivor_head(NULL), + _recorded_survivor_tail(NULL), + _survivors_age_table(true) + { _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime()); _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0; @@ -272,6 +277,15 @@ _concurrent_mark_cleanup_times_ms->add(0.20); _tenuring_threshold = MaxTenuringThreshold; + if (G1UseSurvivorSpace) { + // if G1FixedSurvivorSpaceSize is 0 which means the size is not + // fixed, then _max_survivor_regions will be calculated at + // calculate_young_list_target_config during initialization + _max_survivor_regions = G1FixedSurvivorSpaceSize / HeapRegion::GrainBytes; + } else { + _max_survivor_regions = 0; + } + initialize_all(); } @@ -283,6 +297,9 @@ void G1CollectorPolicy::initialize_flags() { set_min_alignment(HeapRegion::GrainBytes); set_max_alignment(GenRemSet::max_alignment_constraint(rem_set_name())); + if (SurvivorRatio < 1) { + vm_exit_during_initialization("Invalid survivor ratio specified"); + } CollectorPolicy::initialize_flags(); } @@ -301,6 +318,8 @@ "-XX:+UseConcMarkSweepGC."); } + initialize_gc_policy_counters(); + if (G1Gen) { _in_young_gc_mode = true; @@ -322,6 +341,12 @@ } } +// Create the jstat counters for the policy. +void G1CollectorPolicy::initialize_gc_policy_counters() +{ + _gc_policy_counters = new GCPolicyCounters("GarbageFirst", 1, 2 + G1Gen); +} + void G1CollectorPolicy::calculate_young_list_min_length() { _young_list_min_length = 0; @@ -352,6 +377,7 @@ guarantee( so_length < _young_list_target_length, "invariant" ); _young_list_so_prefix_length = so_length; } + calculate_survivors_policy(); } // This method calculate the optimal scan-only set for a fixed young @@ -448,6 +474,9 @@ if (full_young_gcs() && _free_regions_at_end_of_collection > 0) { // we are in fully-young mode and there are free regions in the heap + double survivor_regions_evac_time = + predict_survivor_regions_evac_time(); + size_t min_so_length = 0; size_t max_so_length = 0; @@ -497,9 +526,8 @@ scanned_cards = predict_non_young_card_num(adj_rs_lengths); // calculate this once, so that we don't have to recalculate it in // the innermost loop - double base_time_ms = predict_base_elapsed_time_ms(pending_cards, - scanned_cards); - + double base_time_ms = predict_base_elapsed_time_ms(pending_cards, scanned_cards) + + survivor_regions_evac_time; // the result size_t final_young_length = 0; size_t final_so_length = 0; @@ -548,14 +576,14 @@ bool done = false; // this is the outermost loop while (!done) { -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("searching between " SIZE_FORMAT " and " SIZE_FORMAT ", incr " SIZE_FORMAT ", pass %s", from_so_length, to_so_length, so_length_incr, (pass == pass_type_coarse) ? "coarse" : (pass == pass_type_fine) ? "fine" : "final"); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG size_t so_length = from_so_length; size_t init_free_regions = @@ -651,11 +679,11 @@ guarantee( so_length_incr == so_coarse_increments, "invariant" ); guarantee( final_so_length >= min_so_length, "invariant" ); -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr(" coarse pass: SO length " SIZE_FORMAT, final_so_length); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG from_so_length = (final_so_length - min_so_length > so_coarse_increments) ? @@ -687,12 +715,12 @@ // of the optimal size_t new_so_length = 950 * final_so_length / 1000; -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr(" fine pass: SO length " SIZE_FORMAT ", setting it to " SIZE_FORMAT, final_so_length, new_so_length); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG from_so_length = new_so_length; to_so_length = new_so_length; @@ -719,7 +747,8 @@ } // we should have at least one region in the target young length - _young_list_target_length = MAX2((size_t) 1, final_young_length); + _young_list_target_length = + MAX2((size_t) 1, final_young_length + _recorded_survivor_regions); if (final_so_length >= final_young_length) // and we need to ensure that the S-O length is not greater than // the target young length (this is being a bit careful) @@ -734,7 +763,7 @@ double end_time_sec = os::elapsedTime(); double elapsed_time_ms = (end_time_sec - start_time_sec) * 1000.0; -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("target = %1.1lf ms, young = " SIZE_FORMAT ", SO = " SIZE_FORMAT ", " @@ -747,9 +776,9 @@ calculations, full_young_gcs() ? "full" : "partial", should_initiate_conc_mark() ? " i-m" : "", - in_marking_window(), - in_marking_window_im()); -#endif // 0 + _in_marking_window, + _in_marking_window_im); +#endif // TRACE_CALC_YOUNG_CONFIG if (_young_list_target_length < _young_list_min_length) { // bummer; this means that, if we do a pause when the optimal @@ -768,14 +797,14 @@ // S-O length so_length = calculate_optimal_so_length(_young_list_min_length); -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("adjusted target length from " SIZE_FORMAT " to " SIZE_FORMAT ", SO " SIZE_FORMAT, _young_list_target_length, _young_list_min_length, so_length); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG _young_list_target_length = MAX2(_young_list_min_length, (size_t)1); @@ -785,12 +814,12 @@ // we are in a partially-young mode or we've run out of regions (due // to evacuation failure) -#if 0 +#ifdef TRACE_CALC_YOUNG_CONFIG // leave this in for debugging, just in case gclog_or_tty->print_cr("(partial) setting target to " SIZE_FORMAT ", SO " SIZE_FORMAT, _young_list_min_length, 0); -#endif // 0 +#endif // TRACE_CALC_YOUNG_CONFIG // we'll do the pause as soon as possible and with no S-O prefix // (see above for the reasons behind the latter) @@ -884,6 +913,16 @@ return true; } +double G1CollectorPolicy::predict_survivor_regions_evac_time() { + double survivor_regions_evac_time = 0.0; + for (HeapRegion * r = _recorded_survivor_head; + r != NULL && r != _recorded_survivor_tail->get_next_young_region(); + r = r->get_next_young_region()) { + survivor_regions_evac_time += predict_region_elapsed_time_ms(r, true); + } + return survivor_regions_evac_time; +} + void G1CollectorPolicy::check_prediction_validity() { guarantee( adaptive_young_list_length(), "should not call this otherwise" ); @@ -995,11 +1034,15 @@ _short_lived_surv_rate_group->start_adding_regions(); // also call this on any additional surv rate groups + record_survivor_regions(0, NULL, NULL); + _prev_region_num_young = _region_num_young; _prev_region_num_tenured = _region_num_tenured; _free_regions_at_end_of_collection = _g1->free_regions(); _scan_only_regions_at_end_of_collection = 0; + // Reset survivors SurvRateGroup. + _survivor_surv_rate_group->reset(); calculate_young_list_min_length(); calculate_young_list_target_config(); } @@ -1104,6 +1147,10 @@ _short_lived_surv_rate_group->record_scan_only_prefix(short_lived_so_length); tag_scan_only(short_lived_so_length); + if (G1UseSurvivorSpace) { + _survivors_age_table.clear(); + } + assert( verify_young_ages(), "region age verification" ); } @@ -1965,9 +2012,6 @@ // </NEW PREDICTION> _target_pause_time_ms = -1.0; - - // TODO: calculate tenuring threshold - _tenuring_threshold = MaxTenuringThreshold; } // <NEW PREDICTION> @@ -2058,7 +2102,7 @@ guarantee( hr->is_young() && hr->age_in_surv_rate_group() != -1, "invariant" ); int age = hr->age_in_surv_rate_group(); - double yg_surv_rate = predict_yg_surv_rate(age); + double yg_surv_rate = predict_yg_surv_rate(age, hr->surv_rate_group()); bytes_to_copy = (size_t) ((double) hr->used() * yg_surv_rate); } @@ -2091,7 +2135,7 @@ } #if PREDICTIONS_VERBOSE if (young) { - _recorded_young_bytes += hr->asSpace()->used(); + _recorded_young_bytes += hr->used(); } else { _recorded_marked_bytes += hr->max_live_bytes(); } @@ -2119,11 +2163,6 @@ predict_non_young_card_num(_predicted_rs_lengths); _recorded_region_num = _recorded_young_regions + _recorded_non_young_regions; - _predicted_young_survival_ratio = 0.0; - for (int i = 0; i < _recorded_young_regions; ++i) - _predicted_young_survival_ratio += predict_yg_surv_rate(i); - _predicted_young_survival_ratio /= (double) _recorded_young_regions; - _predicted_scan_only_scan_time_ms = predict_scan_only_time_ms(_recorded_scan_only_regions); _predicted_rs_update_time_ms = @@ -2673,8 +2712,11 @@ assert(in_young_gc_mode(), "should be in young GC mode"); bool ret; size_t young_list_length = _g1->young_list_length(); - - if (young_list_length < _young_list_target_length) { + size_t young_list_max_length = _young_list_target_length; + if (G1FixedEdenSize) { + young_list_max_length -= _max_survivor_regions; + } + if (young_list_length < young_list_max_length) { ret = true; ++_region_num_young; } else { @@ -2710,17 +2752,39 @@ } -uint G1CollectorPolicy::max_regions(int purpose) { +size_t G1CollectorPolicy::max_regions(int purpose) { switch (purpose) { case GCAllocForSurvived: - return G1MaxSurvivorRegions; + return _max_survivor_regions; case GCAllocForTenured: - return UINT_MAX; + return REGIONS_UNLIMITED; default: - return UINT_MAX; + ShouldNotReachHere(); + return REGIONS_UNLIMITED; }; } +// Calculates survivor space parameters. +void G1CollectorPolicy::calculate_survivors_policy() +{ + if (!G1UseSurvivorSpace) { + return; + } + if (G1FixedSurvivorSpaceSize == 0) { + _max_survivor_regions = _young_list_target_length / SurvivorRatio; + } else { + _max_survivor_regions = G1FixedSurvivorSpaceSize / HeapRegion::GrainBytes; + } + + if (G1FixedTenuringThreshold) { + _tenuring_threshold = MaxTenuringThreshold; + } else { + _tenuring_threshold = _survivors_age_table.compute_tenuring_threshold( + HeapRegion::GrainWords * _max_survivor_regions); + } +} + + void G1CollectorPolicy_BestRegionsFirst:: set_single_region_collection_set(HeapRegion* hr) { @@ -2743,7 +2807,11 @@ double max_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0; size_t young_list_length = _g1->young_list_length(); - bool reached_target_length = young_list_length >= _young_list_target_length; + size_t young_list_max_length = _young_list_target_length; + if (G1FixedEdenSize) { + young_list_max_length -= _max_survivor_regions; + } + bool reached_target_length = young_list_length >= young_list_max_length; if (in_young_gc_mode()) { if (reached_target_length) {
--- a/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -49,7 +49,7 @@ class MainBodySummary; class PopPreambleSummary; -class PauseSummary { +class PauseSummary: public CHeapObj { define_num_seq(total) define_num_seq(other) @@ -58,7 +58,7 @@ virtual PopPreambleSummary* pop_preamble_summary() { return NULL; } }; -class MainBodySummary { +class MainBodySummary: public CHeapObj { define_num_seq(satb_drain) // optional define_num_seq(parallel) // parallel only define_num_seq(ext_root_scan) @@ -75,7 +75,7 @@ define_num_seq(clear_ct) // parallel only }; -class PopPreambleSummary { +class PopPreambleSummary: public CHeapObj { define_num_seq(pop_preamble) define_num_seq(pop_update_rs) define_num_seq(pop_scan_rs) @@ -557,6 +557,8 @@ return get_new_neg_prediction(_young_gc_eff_seq); } + double predict_survivor_regions_evac_time(); + // </NEW PREDICTION> public: @@ -599,8 +601,8 @@ // Returns an estimate of the survival rate of the region at yg-age // "yg_age". - double predict_yg_surv_rate(int age) { - TruncatedSeq* seq = _short_lived_surv_rate_group->get_seq(age); + double predict_yg_surv_rate(int age, SurvRateGroup* surv_rate_group) { + TruncatedSeq* seq = surv_rate_group->get_seq(age); if (seq->num() == 0) gclog_or_tty->print("BARF! age is %d", age); guarantee( seq->num() > 0, "invariant" ); @@ -610,6 +612,10 @@ return pred; } + double predict_yg_surv_rate(int age) { + return predict_yg_surv_rate(age, _short_lived_surv_rate_group); + } + double accum_yg_surv_rate_pred(int age) { return _short_lived_surv_rate_group->accum_surv_rate_pred(age); } @@ -822,6 +828,9 @@ virtual void init(); + // Create jstat counters for the policy. + virtual void initialize_gc_policy_counters(); + virtual HeapWord* mem_allocate_work(size_t size, bool is_tlab, bool* gc_overhead_limit_was_exceeded); @@ -1047,8 +1056,12 @@ // Print stats on young survival ratio void print_yg_surv_rate_info() const; - void finished_recalculating_age_indexes() { - _short_lived_surv_rate_group->finished_recalculating_age_indexes(); + void finished_recalculating_age_indexes(bool is_survivors) { + if (is_survivors) { + _survivor_surv_rate_group->finished_recalculating_age_indexes(); + } else { + _short_lived_surv_rate_group->finished_recalculating_age_indexes(); + } // do that for any other surv rate groups } @@ -1097,6 +1110,17 @@ // maximum amount of suvivors regions. int _tenuring_threshold; + // The limit on the number of regions allocated for survivors. + size_t _max_survivor_regions; + + // The amount of survor regions after a collection. + size_t _recorded_survivor_regions; + // List of survivor regions. + HeapRegion* _recorded_survivor_head; + HeapRegion* _recorded_survivor_tail; + + ageTable _survivors_age_table; + public: inline GCAllocPurpose @@ -1116,7 +1140,9 @@ return GCAllocForTenured; } - uint max_regions(int purpose); + static const size_t REGIONS_UNLIMITED = ~(size_t)0; + + size_t max_regions(int purpose); // The limit on regions for a particular purpose is reached. void note_alloc_region_limit_reached(int purpose) { @@ -1132,6 +1158,23 @@ void note_stop_adding_survivor_regions() { _survivor_surv_rate_group->stop_adding_regions(); } + + void record_survivor_regions(size_t regions, + HeapRegion* head, + HeapRegion* tail) { + _recorded_survivor_regions = regions; + _recorded_survivor_head = head; + _recorded_survivor_tail = tail; + } + + void record_thread_age_table(ageTable* age_table) + { + _survivors_age_table.merge_par(age_table); + } + + // Calculates survivor space parameters. + void calculate_survivors_policy(); + }; // This encapsulates a particular strategy for a g1 Collector.
--- a/src/share/vm/gc_implementation/g1/g1MMUTracker.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1MMUTracker.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -28,7 +28,7 @@ /***** ALL TIMES ARE IN SECS!!!!!!! *****/ // this is the "interface" -class G1MMUTracker { +class G1MMUTracker: public CHeapObj { protected: double _time_slice; double _max_gc_time; // this is per time slice @@ -67,7 +67,7 @@ } }; -class G1MMUTrackerQueueElem { +class G1MMUTrackerQueueElem VALUE_OBJ_CLASS_SPEC { private: double _start_time; double _end_time;
--- a/src/share/vm/gc_implementation/g1/g1RemSet.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -572,6 +572,9 @@ } guarantee( _cards_scanned == NULL, "invariant" ); _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); + for (uint i = 0; i < n_workers(); ++i) { + _cards_scanned[i] = 0; + } _total_cards_scanned = 0; }
--- a/src/share/vm/gc_implementation/g1/g1RemSet.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1RemSet.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -30,7 +30,7 @@ class HRInto_G1RemSet; class ConcurrentG1Refine; -class G1RemSet { +class G1RemSet: public CHeapObj { protected: G1CollectedHeap* _g1;
--- a/src/share/vm/gc_implementation/g1/g1_globals.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/g1_globals.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -281,7 +281,17 @@ develop(bool, G1HRRSFlushLogBuffersOnVerify, false, \ "Forces flushing of log buffers before verification.") \ \ - product(intx, G1MaxSurvivorRegions, 0, \ - "The maximum number of survivor regions") + product(bool, G1UseSurvivorSpace, true, \ + "When true, use survivor space.") \ + \ + product(bool, G1FixedTenuringThreshold, false, \ + "When set, G1 will not adjust the tenuring threshold") \ + \ + product(bool, G1FixedEdenSize, false, \ + "When set, G1 will not allocate unused survivor space regions") \ + \ + product(uintx, G1FixedSurvivorSpaceSize, 0, \ + "If non-0 is the size of the G1 survivor space, " \ + "otherwise SurvivorRatio is used to determine the size") G1_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_EXPERIMENTAL_FLAG, DECLARE_NOTPRODUCT_FLAG, DECLARE_MANAGEABLE_FLAG, DECLARE_PRODUCT_RW_FLAG)
--- a/src/share/vm/gc_implementation/g1/heapRegion.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/heapRegion.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -566,7 +566,11 @@ void note_end_of_copying() { assert(top() >= _next_top_at_mark_start, "Increase only"); - _next_top_at_mark_start = top(); + // Survivor regions will be scanned on the start of concurrent + // marking. + if (!is_survivor()) { + _next_top_at_mark_start = top(); + } } // Returns "false" iff no object in the region was allocated when the @@ -829,7 +833,7 @@ // A linked lists of heap regions. It leaves the "next" field // unspecified; that's up to subtypes. -class RegionList { +class RegionList VALUE_OBJ_CLASS_SPEC { protected: virtual HeapRegion* get_next(HeapRegion* chr) = 0; virtual void set_next(HeapRegion* chr,
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -65,9 +65,11 @@ // We need access in order to union things into the base table. BitMap* bm() { return &_bm; } +#if PRT_COUNT_OCCUPIED void recount_occupied() { _occupied = (jint) bm()->count_one_bits(); } +#endif PerRegionTable(HeapRegion* hr) : _hr(hr), @@ -1144,7 +1146,9 @@ size_t i = _outgoing_region_map.get_next_one_offset(0); while (i < _outgoing_region_map.size()) { HeapRegion* to_region = g1h->region_at(i); - to_region->rem_set()->clear_incoming_entry(hr()); + if (!to_region->in_collection_set()) { + to_region->rem_set()->clear_incoming_entry(hr()); + } i = _outgoing_region_map.get_next_one_offset(i+1); } }
--- a/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -58,7 +58,7 @@ // is represented. If a deleted PRT is re-used, a thread adding a bit, // thinking the PRT is for a different region, does no harm. -class OtherRegionsTable: public CHeapObj { +class OtherRegionsTable VALUE_OBJ_CLASS_SPEC { friend class HeapRegionRemSetIterator; G1CollectedHeap* _g1h;
--- a/src/share/vm/gc_implementation/g1/ptrQueue.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/ptrQueue.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -29,7 +29,7 @@ class PtrQueueSet; -class PtrQueue: public CHeapObj { +class PtrQueue VALUE_OBJ_CLASS_SPEC { protected: // The ptr queue set to which this queue belongs. @@ -130,7 +130,7 @@ // In particular, the individual queues allocate buffers from this shared // set, and return completed buffers to the set. // All these variables are are protected by the TLOQ_CBL_mon. XXX ??? -class PtrQueueSet: public CHeapObj { +class PtrQueueSet VALUE_OBJ_CLASS_SPEC { protected:
--- a/src/share/vm/gc_implementation/g1/sparsePRT.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/sparsePRT.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -33,7 +33,7 @@ // old versions synchronously. -class SparsePRTEntry { +class SparsePRTEntry: public CHeapObj { public: enum SomePublicConstants { CardsPerEntry = (short)4, @@ -167,7 +167,7 @@ }; // ValueObj because will be embedded in HRRS iterator. -class RSHashTableIter: public CHeapObj { +class RSHashTableIter VALUE_OBJ_CLASS_SPEC { short _tbl_ind; short _bl_ind; short _card_ind; @@ -213,7 +213,7 @@ class SparsePRTIter; -class SparsePRT : public CHeapObj { +class SparsePRT VALUE_OBJ_CLASS_SPEC { // Iterations are done on the _cur hash table, since they only need to // see entries visible at the start of a collection pause. // All other operations are done using the _next hash table.
--- a/src/share/vm/gc_implementation/g1/survRateGroup.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/survRateGroup.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -29,23 +29,14 @@ const char* name, size_t summary_surv_rates_len) : _g1p(g1p), _name(name), - _all_regions_allocated(0), - _curr_length(0), _scan_only_prefix(0), _setup_seq_num(0), - _array_length(0), _surv_rate(NULL), _accum_surv_rate_pred(NULL), - _accum_surv_rate(0.0), _surv_rate_pred(NULL), _last_pred(0.0), _summary_surv_rates_len(summary_surv_rates_len), _summary_surv_rates_max_len(0), - _summary_surv_rates(NULL) { - - // the following will set up the arrays with length 1 - _curr_length = 1; - stop_adding_regions(); - guarantee( _array_length == 1, "invariant" ); - guarantee( _surv_rate_pred[0] != NULL, "invariant" ); - _surv_rate_pred[0]->add(0.4); - all_surviving_words_recorded(false); - _curr_length = 0; - + _summary_surv_rates(NULL), + _surv_rate(NULL), + _accum_surv_rate_pred(NULL), + _surv_rate_pred(NULL) +{ + reset(); if (summary_surv_rates_len > 0) { size_t length = summary_surv_rates_len; _summary_surv_rates = NEW_C_HEAP_ARRAY(NumberSeq*, length); @@ -60,61 +51,80 @@ start_adding_regions(); } + +void SurvRateGroup::reset() +{ + _all_regions_allocated = 0; + _scan_only_prefix = 0; + _setup_seq_num = 0; + _stats_arrays_length = 0; + _accum_surv_rate = 0.0; + _last_pred = 0.0; + // the following will set up the arrays with length 1 + _region_num = 1; + stop_adding_regions(); + guarantee( _stats_arrays_length == 1, "invariant" ); + guarantee( _surv_rate_pred[0] != NULL, "invariant" ); + _surv_rate_pred[0]->add(0.4); + all_surviving_words_recorded(false); + _region_num = 0; +} + + void SurvRateGroup::start_adding_regions() { - _setup_seq_num = _array_length; - _curr_length = _scan_only_prefix; + _setup_seq_num = _stats_arrays_length; + _region_num = _scan_only_prefix; _accum_surv_rate = 0.0; #if 0 - gclog_or_tty->print_cr("start adding regions, seq num %d, length %d", - _setup_seq_num, _curr_length); + gclog_or_tty->print_cr("[%s] start adding regions, seq num %d, length %d", + _name, _setup_seq_num, _region_num); #endif // 0 } void SurvRateGroup::stop_adding_regions() { - size_t length = _curr_length; #if 0 - gclog_or_tty->print_cr("stop adding regions, length %d", length); + gclog_or_tty->print_cr("[%s] stop adding regions, length %d", _name, _region_num); #endif // 0 - if (length > _array_length) { + if (_region_num > _stats_arrays_length) { double* old_surv_rate = _surv_rate; double* old_accum_surv_rate_pred = _accum_surv_rate_pred; TruncatedSeq** old_surv_rate_pred = _surv_rate_pred; - _surv_rate = NEW_C_HEAP_ARRAY(double, length); + _surv_rate = NEW_C_HEAP_ARRAY(double, _region_num); if (_surv_rate == NULL) { - vm_exit_out_of_memory(sizeof(double) * length, + vm_exit_out_of_memory(sizeof(double) * _region_num, "Not enough space for surv rate array."); } - _accum_surv_rate_pred = NEW_C_HEAP_ARRAY(double, length); + _accum_surv_rate_pred = NEW_C_HEAP_ARRAY(double, _region_num); if (_accum_surv_rate_pred == NULL) { - vm_exit_out_of_memory(sizeof(double) * length, + vm_exit_out_of_memory(sizeof(double) * _region_num, "Not enough space for accum surv rate pred array."); } - _surv_rate_pred = NEW_C_HEAP_ARRAY(TruncatedSeq*, length); + _surv_rate_pred = NEW_C_HEAP_ARRAY(TruncatedSeq*, _region_num); if (_surv_rate == NULL) { - vm_exit_out_of_memory(sizeof(TruncatedSeq*) * length, + vm_exit_out_of_memory(sizeof(TruncatedSeq*) * _region_num, "Not enough space for surv rate pred array."); } - for (size_t i = 0; i < _array_length; ++i) + for (size_t i = 0; i < _stats_arrays_length; ++i) _surv_rate_pred[i] = old_surv_rate_pred[i]; #if 0 - gclog_or_tty->print_cr("stop adding regions, new seqs %d to %d", - _array_length, length - 1); + gclog_or_tty->print_cr("[%s] stop adding regions, new seqs %d to %d", + _name, _array_length, _region_num - 1); #endif // 0 - for (size_t i = _array_length; i < length; ++i) { + for (size_t i = _stats_arrays_length; i < _region_num; ++i) { _surv_rate_pred[i] = new TruncatedSeq(10); // _surv_rate_pred[i]->add(last_pred); } - _array_length = length; + _stats_arrays_length = _region_num; if (old_surv_rate != NULL) FREE_C_HEAP_ARRAY(double, old_surv_rate); @@ -124,7 +134,7 @@ FREE_C_HEAP_ARRAY(NumberSeq*, old_surv_rate_pred); } - for (size_t i = 0; i < _array_length; ++i) + for (size_t i = 0; i < _stats_arrays_length; ++i) _surv_rate[i] = 0.0; } @@ -135,7 +145,7 @@ double ret = _accum_surv_rate; if (adjustment > 0) { - TruncatedSeq* seq = get_seq(_curr_length+1); + TruncatedSeq* seq = get_seq(_region_num+1); double surv_rate = _g1p->get_new_prediction(seq); ret += surv_rate; } @@ -145,23 +155,23 @@ int SurvRateGroup::next_age_index() { - TruncatedSeq* seq = get_seq(_curr_length); + TruncatedSeq* seq = get_seq(_region_num); double surv_rate = _g1p->get_new_prediction(seq); _accum_surv_rate += surv_rate; - ++_curr_length; + ++_region_num; return (int) ++_all_regions_allocated; } void SurvRateGroup::record_scan_only_prefix(size_t scan_only_prefix) { - guarantee( scan_only_prefix <= _curr_length, "pre-condition" ); + guarantee( scan_only_prefix <= _region_num, "pre-condition" ); _scan_only_prefix = scan_only_prefix; } void SurvRateGroup::record_surviving_words(int age_in_group, size_t surv_words) { - guarantee( 0 <= age_in_group && (size_t) age_in_group < _curr_length, + guarantee( 0 <= age_in_group && (size_t) age_in_group < _region_num, "pre-condition" ); guarantee( _surv_rate[age_in_group] <= 0.00001, "should only update each slot once" ); @@ -178,15 +188,15 @@ void SurvRateGroup::all_surviving_words_recorded(bool propagate) { - if (propagate && _curr_length > 0) { // conservative - double surv_rate = _surv_rate_pred[_curr_length-1]->last(); + if (propagate && _region_num > 0) { // conservative + double surv_rate = _surv_rate_pred[_region_num-1]->last(); #if 0 gclog_or_tty->print_cr("propagating %1.2lf from %d to %d", surv_rate, _curr_length, _array_length - 1); #endif // 0 - for (size_t i = _curr_length; i < _array_length; ++i) { + for (size_t i = _region_num; i < _stats_arrays_length; ++i) { guarantee( _surv_rate[i] <= 0.00001, "the slot should not have been updated" ); _surv_rate_pred[i]->add(surv_rate); @@ -195,7 +205,7 @@ double accum = 0.0; double pred = 0.0; - for (size_t i = 0; i < _array_length; ++i) { + for (size_t i = 0; i < _stats_arrays_length; ++i) { pred = _g1p->get_new_prediction(_surv_rate_pred[i]); if (pred > 1.0) pred = 1.0; accum += pred; @@ -209,8 +219,8 @@ void SurvRateGroup::print() { gclog_or_tty->print_cr("Surv Rate Group: %s (%d entries, %d scan-only)", - _name, _curr_length, _scan_only_prefix); - for (size_t i = 0; i < _curr_length; ++i) { + _name, _region_num, _scan_only_prefix); + for (size_t i = 0; i < _region_num; ++i) { gclog_or_tty->print_cr(" age %4d surv rate %6.2lf %% pred %6.2lf %%%s", i, _surv_rate[i] * 100.0, _g1p->get_new_prediction(_surv_rate_pred[i]) * 100.0,
--- a/src/share/vm/gc_implementation/g1/survRateGroup.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/g1/survRateGroup.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -29,7 +29,7 @@ G1CollectorPolicy* _g1p; const char* _name; - size_t _array_length; + size_t _stats_arrays_length; double* _surv_rate; double* _accum_surv_rate_pred; double _last_pred; @@ -40,7 +40,7 @@ size_t _summary_surv_rates_max_len; int _all_regions_allocated; - size_t _curr_length; + size_t _region_num; size_t _scan_only_prefix; size_t _setup_seq_num; @@ -48,6 +48,7 @@ SurvRateGroup(G1CollectorPolicy* g1p, const char* name, size_t summary_surv_rates_len); + void reset(); void start_adding_regions(); void stop_adding_regions(); void record_scan_only_prefix(size_t scan_only_prefix); @@ -55,22 +56,21 @@ void all_surviving_words_recorded(bool propagate); const char* name() { return _name; } - size_t region_num() { return _curr_length; } + size_t region_num() { return _region_num; } size_t scan_only_length() { return _scan_only_prefix; } double accum_surv_rate_pred(int age) { assert(age >= 0, "must be"); - if ((size_t)age < _array_length) + if ((size_t)age < _stats_arrays_length) return _accum_surv_rate_pred[age]; else { - double diff = (double) (age - _array_length + 1); - return _accum_surv_rate_pred[_array_length-1] + diff * _last_pred; + double diff = (double) (age - _stats_arrays_length + 1); + return _accum_surv_rate_pred[_stats_arrays_length-1] + diff * _last_pred; } } double accum_surv_rate(size_t adjustment); TruncatedSeq* get_seq(size_t age) { - guarantee( 0 <= age, "pre-condition" ); if (age >= _setup_seq_num) { guarantee( _setup_seq_num > 0, "invariant" ); age = _setup_seq_num-1;
--- a/src/share/vm/gc_implementation/includeDB_gc_g1 Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/includeDB_gc_g1 Wed Feb 25 23:16:22 2009 -0800 @@ -48,6 +48,7 @@ concurrentG1Refine.cpp space.inline.hpp concurrentG1Refine.hpp globalDefinitions.hpp +concurrentG1Refine.hpp allocation.hpp concurrentG1RefineThread.cpp concurrentG1Refine.hpp concurrentG1RefineThread.cpp concurrentG1RefineThread.hpp @@ -172,6 +173,7 @@ g1CollectorPolicy.cpp g1CollectorPolicy.hpp g1CollectorPolicy.cpp heapRegionRemSet.hpp g1CollectorPolicy.cpp mutexLocker.hpp +g1CollectorPolicy.cpp gcPolicyCounters.hpp g1CollectorPolicy.hpp collectorPolicy.hpp g1CollectorPolicy.hpp collectionSetChooser.hpp @@ -228,7 +230,7 @@ g1MMUTracker.cpp mutexLocker.hpp g1MMUTracker.hpp debug.hpp - +g1MMUTracker.hpp allocation.hpp g1RemSet.cpp bufferingOopClosure.hpp g1RemSet.cpp concurrentG1Refine.hpp g1RemSet.cpp concurrentG1RefineThread.hpp @@ -272,6 +274,7 @@ heapRegion.hpp watermark.hpp heapRegion.hpp g1_specialized_oop_closures.hpp heapRegion.hpp survRateGroup.hpp +heapRegion.hpp ageTable.hpp heapRegionRemSet.hpp sparsePRT.hpp
--- a/src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -362,6 +362,10 @@ if (PrintHeapAtGC) { Universe::print_heap_after_gc(); } + +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif } bool PSMarkSweep::absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
--- a/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -2203,6 +2203,10 @@ collection_exit.ticks()); gc_task_manager()->print_task_time_stamps(); } + +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif } bool PSParallelCompact::absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
--- a/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/parallelScavenge/psScavenge.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -615,6 +615,10 @@ gc_task_manager()->print_task_time_stamps(); } +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif + return !promotion_failure_occurred; }
--- a/src/share/vm/gc_implementation/shared/ageTable.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/ageTable.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -67,6 +67,12 @@ } } +void ageTable::merge_par(ageTable* subTable) { + for (int i = 0; i < table_size; i++) { + Atomic::add_ptr(subTable->sizes[i], &sizes[i]); + } +} + int ageTable::compute_tenuring_threshold(size_t survivor_capacity) { size_t desired_survivor_size = (size_t)((((double) survivor_capacity)*TargetSurvivorRatio)/100); size_t total = 0;
--- a/src/share/vm/gc_implementation/shared/ageTable.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_implementation/shared/ageTable.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -56,6 +56,7 @@ // Merge another age table with the current one. Used // for parallel young generation gc. void merge(ageTable* subTable); + void merge_par(ageTable* subTable); // calculate new tenuring threshold based on age information int compute_tenuring_threshold(size_t survivor_capacity);
--- a/src/share/vm/gc_interface/collectedHeap.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/gc_interface/collectedHeap.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -42,6 +42,7 @@ class CollectedHeap : public CHeapObj { friend class VMStructs; friend class IsGCActiveMark; // Block structured external access to _is_gc_active + friend class constantPoolCacheKlass; // allocate() method inserts is_conc_safe #ifdef ASSERT static int _fire_out_of_memory_count; @@ -82,8 +83,6 @@ // Reinitialize tlabs before resuming mutators. virtual void resize_all_tlabs(); - debug_only(static void check_for_valid_allocation_state();) - protected: // Allocate from the current thread's TLAB, with broken-out slow path. inline static HeapWord* allocate_from_tlab(Thread* thread, size_t size); @@ -142,6 +141,7 @@ PRODUCT_RETURN; virtual void check_for_non_bad_heap_word_value(HeapWord* addr, size_t size) PRODUCT_RETURN; + debug_only(static void check_for_valid_allocation_state();) public: enum Name {
--- a/src/share/vm/interpreter/rewriter.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/interpreter/rewriter.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -48,9 +48,14 @@ // Creates a constant pool cache given an inverse_index_map +// This creates the constant pool cache initially in a state +// that is unsafe for concurrent GC processing but sets it to +// a safe mode before the constant pool cache is returned. constantPoolCacheHandle Rewriter::new_constant_pool_cache(intArray& inverse_index_map, TRAPS) { const int length = inverse_index_map.length(); - constantPoolCacheOop cache = oopFactory::new_constantPoolCache(length, CHECK_(constantPoolCacheHandle())); + constantPoolCacheOop cache = oopFactory::new_constantPoolCache(length, + methodOopDesc::IsUnsafeConc, + CHECK_(constantPoolCacheHandle())); cache->initialize(inverse_index_map); return constantPoolCacheHandle(THREAD, cache); }
--- a/src/share/vm/libadt/dict.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/libadt/dict.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -346,9 +346,12 @@ return strcmp((const char *)k1,(const char *)k2); } -// Slimey cheap key comparator. +// Cheap key comparator. int32 cmpkey(const void *key1, const void *key2) { - return (int32)((intptr_t)key1 - (intptr_t)key2); + if (key1 == key2) return 0; + intptr_t delta = (intptr_t)key1 - (intptr_t)key2; + if (delta > 0) return 1; + return -1; } //=============================================================================
--- a/src/share/vm/memory/genCollectedHeap.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/memory/genCollectedHeap.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -610,6 +610,10 @@ Universe::print_heap_after_gc(); } +#ifdef TRACESPINNING + ParallelTaskTerminator::print_termination_counts(); +#endif + if (ExitAfterGCNum > 0 && total_collections() == ExitAfterGCNum) { tty->print_cr("Stopping after GC #%d", ExitAfterGCNum); vm_exit(-1);
--- a/src/share/vm/memory/oopFactory.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/memory/oopFactory.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -90,9 +90,11 @@ } -constantPoolCacheOop oopFactory::new_constantPoolCache(int length, TRAPS) { +constantPoolCacheOop oopFactory::new_constantPoolCache(int length, + bool is_conc_safe, + TRAPS) { constantPoolCacheKlass* ck = constantPoolCacheKlass::cast(Universe::constantPoolCacheKlassObj()); - return ck->allocate(length, CHECK_NULL); + return ck->allocate(length, is_conc_safe, CHECK_NULL); }
--- a/src/share/vm/memory/oopFactory.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/memory/oopFactory.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -84,7 +84,9 @@ static constantPoolOop new_constantPool (int length, bool is_conc_safe, TRAPS); - static constantPoolCacheOop new_constantPoolCache(int length, TRAPS); + static constantPoolCacheOop new_constantPoolCache(int length, + bool is_conc_safe, + TRAPS); // Instance classes static klassOop new_instanceKlass(int vtable_len, int itable_len, int static_field_size,
--- a/src/share/vm/oops/cpCacheKlass.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/oops/cpCacheKlass.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -32,13 +32,43 @@ } -constantPoolCacheOop constantPoolCacheKlass::allocate(int length, TRAPS) { +constantPoolCacheOop constantPoolCacheKlass::allocate(int length, + bool is_conc_safe, + TRAPS) { // allocate memory int size = constantPoolCacheOopDesc::object_size(length); + KlassHandle klass (THREAD, as_klassOop()); - constantPoolCacheOop cache = (constantPoolCacheOop) - CollectedHeap::permanent_obj_allocate(klass, size, CHECK_NULL); + + // This is the original code. The code from permanent_obj_allocate() + // was in-lined to allow the setting of is_conc_safe before the klass + // is installed. + // constantPoolCacheOop cache = (constantPoolCacheOop) + // CollectedHeap::permanent_obj_allocate(klass, size, CHECK_NULL); + + oop obj = CollectedHeap::permanent_obj_allocate_no_klass_install(klass, size, CHECK_NULL); + constantPoolCacheOop cache = (constantPoolCacheOop) obj; + cache->set_is_conc_safe(is_conc_safe); + // The store to is_conc_safe must be visible before the klass + // is set. This should be done safely because _is_conc_safe has + // been declared volatile. If there are any problems, consider adding + // OrderAccess::storestore(); + CollectedHeap::post_allocation_install_obj_klass(klass, obj, size); + NOT_PRODUCT(Universe::heap()->check_for_bad_heap_word_value((HeapWord*) obj, + size)); + + // The length field affects the size of the object. The allocation + // above allocates the correct size (see calculation of "size") but + // the size() method of the constant pool cache oop will not reflect + // that size until the correct length is set. cache->set_length(length); + + // The store of the length must be visible before is_conc_safe is + // set to a safe state. + // This should be done safely because _is_conc_safe has + // been declared volatile. If there are any problems, consider adding + // OrderAccess::storestore(); + cache->set_is_conc_safe(methodOopDesc::IsSafeConc); cache->set_constant_pool(NULL); return cache; } @@ -114,7 +144,6 @@ return size; } - int constantPoolCacheKlass::oop_adjust_pointers(oop obj) { assert(obj->is_constantPoolCache(), "obj must be constant pool cache"); constantPoolCacheOop cache = (constantPoolCacheOop)obj; @@ -131,6 +160,11 @@ return size; } +bool constantPoolCacheKlass::oop_is_conc_safe(oop obj) const { + assert(obj->is_constantPoolCache(), "should be constant pool"); + return constantPoolCacheOop(obj)->is_conc_safe(); +} + #ifndef SERIALGC void constantPoolCacheKlass::oop_copy_contents(PSPromotionManager* pm, oop obj) {
--- a/src/share/vm/oops/cpCacheKlass.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/oops/cpCacheKlass.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -32,7 +32,7 @@ // Allocation DEFINE_ALLOCATE_PERMANENT(constantPoolCacheKlass); - constantPoolCacheOop allocate(int length, TRAPS); + constantPoolCacheOop allocate(int length, bool is_conc_safe, TRAPS); static klassOop create_klass(TRAPS); // Casting from klassOop @@ -48,6 +48,7 @@ // Garbage collection void oop_follow_contents(oop obj); int oop_adjust_pointers(oop obj); + virtual bool oop_is_conc_safe(oop obj) const; // Parallel Scavenge and Parallel Old PARALLEL_GC_DECLS
--- a/src/share/vm/oops/cpCacheOop.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/oops/cpCacheOop.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -291,6 +291,9 @@ private: int _length; constantPoolOop _constant_pool; // the corresponding constant pool + // If true, safe for concurrent GC processing, + // Set unconditionally in constantPoolCacheKlass::allocate() + volatile bool _is_conc_safe; // Sizing debug_only(friend class ClassVerifier;) @@ -316,6 +319,12 @@ constantPoolOop constant_pool() const { return _constant_pool; } ConstantPoolCacheEntry* entry_at(int i) const { assert(0 <= i && i < length(), "index out of bounds"); return base() + i; } + // GC support + // If the _length field has not been set, the size of the + // constantPoolCache cannot be correctly calculated. + bool is_conc_safe() { return _is_conc_safe; } + void set_is_conc_safe(bool v) { _is_conc_safe = v; } + // Code generation static ByteSize base_offset() { return in_ByteSize(sizeof(constantPoolCacheOopDesc)); }
--- a/src/share/vm/opto/block.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/block.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -880,6 +880,7 @@ } void PhaseCFG::verify( ) const { +#ifdef ASSERT // Verify sane CFG for( uint i = 0; i < _num_blocks; i++ ) { Block *b = _blocks[i]; @@ -894,10 +895,20 @@ "CreateEx must be first instruction in block" ); } for( uint k = 0; k < n->req(); k++ ) { - Node *use = n->in(k); - if( use && use != n ) { - assert( _bbs[use->_idx] || use->is_Con(), + Node *def = n->in(k); + if( def && def != n ) { + assert( _bbs[def->_idx] || def->is_Con(), "must have block; constants for debug info ok" ); + // Verify that instructions in the block is in correct order. + // Uses must follow their definition if they are at the same block. + // Mostly done to check that MachSpillCopy nodes are placed correctly + // when CreateEx node is moved in build_ifg_physical(). + if( _bbs[def->_idx] == b && + !(b->head()->is_Loop() && n->is_Phi()) && + // See (+++) comment in reg_split.cpp + !(n->jvms() != NULL && n->jvms()->is_monitor_use(k)) ) { + assert( b->find_node(def) < j, "uses must follow definitions" ); + } } } } @@ -914,6 +925,7 @@ assert( b->_num_succs == 2, "Conditional branch must have two targets"); } } +#endif } #endif
--- a/src/share/vm/opto/c2_globals.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/c2_globals.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -191,6 +191,9 @@ notproduct(bool, VerifyHashTableKeys, true, \ "Verify the immutability of keys in the VN hash tables") \ \ + notproduct(bool, VerifyRegisterAllocator , false, \ + "Verify Register Allocator") \ + \ develop_pd(intx, FLOATPRESSURE, \ "Number of float LRG's that constitute high register pressure") \ \
--- a/src/share/vm/opto/cfgnode.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/cfgnode.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -858,12 +858,18 @@ // convert the one to the other. const TypePtr* ttp = _type->make_ptr(); const TypeInstPtr* ttip = (ttp != NULL) ? ttp->isa_instptr() : NULL; + const TypeKlassPtr* ttkp = (ttp != NULL) ? ttp->isa_klassptr() : NULL; bool is_intf = false; if (ttip != NULL) { ciKlass* k = ttip->klass(); if (k->is_loaded() && k->is_interface()) is_intf = true; } + if (ttkp != NULL) { + ciKlass* k = ttkp->klass(); + if (k->is_loaded() && k->is_interface()) + is_intf = true; + } // Default case: merge all inputs const Type *t = Type::TOP; // Merged type starting value @@ -921,6 +927,8 @@ // uplift the type. if( !t->empty() && ttip && ttip->is_loaded() && ttip->klass()->is_interface() ) { assert(ft == _type, ""); } // Uplift to interface + else if( !t->empty() && ttkp && ttkp->is_loaded() && ttkp->klass()->is_interface() ) + { assert(ft == _type, ""); } // Uplift to interface // Otherwise it's something stupid like non-overlapping int ranges // found on dying counted loops. else @@ -936,6 +944,7 @@ // because the type system doesn't interact well with interfaces. const TypePtr *jtp = jt->make_ptr(); const TypeInstPtr *jtip = (jtp != NULL) ? jtp->isa_instptr() : NULL; + const TypeKlassPtr *jtkp = (jtp != NULL) ? jtp->isa_klassptr() : NULL; if( jtip && ttip ) { if( jtip->is_loaded() && jtip->klass()->is_interface() && ttip->is_loaded() && !ttip->klass()->is_interface() ) { @@ -945,6 +954,14 @@ jt = ft; } } + if( jtkp && ttkp ) { + if( jtkp->is_loaded() && jtkp->klass()->is_interface() && + ttkp->is_loaded() && !ttkp->klass()->is_interface() ) { + assert(ft == ttkp->cast_to_ptr_type(jtkp->ptr()) || + ft->isa_narrowoop() && ft->make_ptr() == ttkp->cast_to_ptr_type(jtkp->ptr()), ""); + jt = ft; + } + } if (jt != ft && jt->base() == ft->base()) { if (jt->isa_int() && jt->is_int()->_lo == ft->is_int()->_lo &&
--- a/src/share/vm/opto/chaitin.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/chaitin.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -228,6 +228,11 @@ // them for real. de_ssa(); +#ifdef ASSERT + // Veify the graph before RA. + verify(&live_arena); +#endif + { NOT_PRODUCT( Compile::TracePhase t3("computeLive", &_t_computeLive, TimeCompiler); ) _live = NULL; // Mark live as being not available @@ -306,12 +311,6 @@ C->check_node_count(2*NodeLimitFudgeFactor, "out of nodes after physical split"); if (C->failing()) return; -#ifdef ASSERT - if( VerifyOpto ) { - _cfg.verify(); - verify_base_ptrs(&live_arena); - } -#endif NOT_PRODUCT( C->verify_graph_edges(); ) compact(); // Compact LRGs; return new lower max lrg @@ -340,7 +339,7 @@ compress_uf_map_for_nodes(); #ifdef ASSERT - if( VerifyOpto ) _ifg->verify(this); + verify(&live_arena, true); #endif } else { ifg.SquareUp(); @@ -376,12 +375,6 @@ // Bail out if unique gets too large (ie - unique > MaxNodeLimit - 2*NodeLimitFudgeFactor) C->check_node_count(2*NodeLimitFudgeFactor, "out of nodes after split"); if (C->failing()) return; -#ifdef ASSERT - if( VerifyOpto ) { - _cfg.verify(); - verify_base_ptrs(&live_arena); - } -#endif compact(); // Compact LRGs; return new lower max lrg @@ -412,7 +405,7 @@ } compress_uf_map_for_nodes(); #ifdef ASSERT - if( VerifyOpto ) _ifg->verify(this); + verify(&live_arena, true); #endif cache_lrg_info(); // Count degree of LRGs @@ -432,6 +425,11 @@ // Peephole remove copies post_allocate_copy_removal(); +#ifdef ASSERT + // Veify the graph after RA. + verify(&live_arena); +#endif + // max_reg is past the largest *register* used. // Convert that to a frame_slot number. if( _max_reg <= _matcher._new_SP ) @@ -956,7 +954,7 @@ while ((neighbor = elements.next()) != 0) { LRG *n = &lrgs(neighbor); #ifdef ASSERT - if( VerifyOpto ) { + if( VerifyOpto || VerifyRegisterAllocator ) { assert( _ifg->effective_degree(neighbor) == n->degree(), "" ); } #endif
--- a/src/share/vm/opto/chaitin.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/chaitin.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -491,6 +491,8 @@ // Verify that base pointers and derived pointers are still sane void verify_base_ptrs( ResourceArea *a ) const; + void verify( ResourceArea *a, bool verify_ifg = false ) const; + void dump_for_spill_split_recycle() const; public:
--- a/src/share/vm/opto/classes.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/classes.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -129,7 +129,7 @@ macro(LShiftI) macro(LShiftL) macro(LoadB) -macro(LoadC) +macro(LoadUS) macro(LoadD) macro(LoadD_unaligned) macro(LoadF)
--- a/src/share/vm/opto/compile.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/compile.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -2005,7 +2005,7 @@ case Op_StoreP: case Op_StoreN: case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadI: case Op_LoadKlass: case Op_LoadNKlass:
--- a/src/share/vm/opto/divnode.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/divnode.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 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 @@ -244,42 +244,73 @@ //---------------------long_by_long_mulhi-------------------------------------- // Generate ideal node graph for upper half of a 64 bit x 64 bit multiplication -static Node *long_by_long_mulhi( PhaseGVN *phase, Node *dividend, jlong magic_const) { +static Node* long_by_long_mulhi(PhaseGVN* phase, Node* dividend, jlong magic_const) { // If the architecture supports a 64x64 mulhi, there is // no need to synthesize it in ideal nodes. if (Matcher::has_match_rule(Op_MulHiL)) { - Node *v = phase->longcon(magic_const); + Node* v = phase->longcon(magic_const); return new (phase->C, 3) MulHiLNode(dividend, v); } + // Taken from Hacker's Delight, Fig. 8-2. Multiply high signed. + // (http://www.hackersdelight.org/HDcode/mulhs.c) + // + // int mulhs(int u, int v) { + // unsigned u0, v0, w0; + // int u1, v1, w1, w2, t; + // + // u0 = u & 0xFFFF; u1 = u >> 16; + // v0 = v & 0xFFFF; v1 = v >> 16; + // w0 = u0*v0; + // t = u1*v0 + (w0 >> 16); + // w1 = t & 0xFFFF; + // w2 = t >> 16; + // w1 = u0*v1 + w1; + // return u1*v1 + w2 + (w1 >> 16); + // } + // + // Note: The version above is for 32x32 multiplications, while the + // following inline comments are adapted to 64x64. + const int N = 64; - Node *u_hi = phase->transform(new (phase->C, 3) RShiftLNode(dividend, phase->intcon(N / 2))); - Node *u_lo = phase->transform(new (phase->C, 3) AndLNode(dividend, phase->longcon(0xFFFFFFFF))); + // u0 = u & 0xFFFFFFFF; u1 = u >> 32; + Node* u0 = phase->transform(new (phase->C, 3) AndLNode(dividend, phase->longcon(0xFFFFFFFF))); + Node* u1 = phase->transform(new (phase->C, 3) RShiftLNode(dividend, phase->intcon(N / 2))); + + // v0 = v & 0xFFFFFFFF; v1 = v >> 32; + Node* v0 = phase->longcon(magic_const & 0xFFFFFFFF); + Node* v1 = phase->longcon(magic_const >> (N / 2)); - Node *v_hi = phase->longcon(magic_const >> N/2); - Node *v_lo = phase->longcon(magic_const & 0XFFFFFFFF); + // w0 = u0*v0; + Node* w0 = phase->transform(new (phase->C, 3) MulLNode(u0, v0)); - Node *hihi_product = phase->transform(new (phase->C, 3) MulLNode(u_hi, v_hi)); - Node *hilo_product = phase->transform(new (phase->C, 3) MulLNode(u_hi, v_lo)); - Node *lohi_product = phase->transform(new (phase->C, 3) MulLNode(u_lo, v_hi)); - Node *lolo_product = phase->transform(new (phase->C, 3) MulLNode(u_lo, v_lo)); + // t = u1*v0 + (w0 >> 32); + Node* u1v0 = phase->transform(new (phase->C, 3) MulLNode(u1, v0)); + Node* temp = phase->transform(new (phase->C, 3) URShiftLNode(w0, phase->intcon(N / 2))); + Node* t = phase->transform(new (phase->C, 3) AddLNode(u1v0, temp)); + + // w1 = t & 0xFFFFFFFF; + Node* w1 = new (phase->C, 3) AndLNode(t, phase->longcon(0xFFFFFFFF)); - Node *t1 = phase->transform(new (phase->C, 3) URShiftLNode(lolo_product, phase->intcon(N / 2))); - Node *t2 = phase->transform(new (phase->C, 3) AddLNode(hilo_product, t1)); + // w2 = t >> 32; + Node* w2 = new (phase->C, 3) RShiftLNode(t, phase->intcon(N / 2)); + + // 6732154: Construct both w1 and w2 before transforming, so t + // doesn't go dead prematurely. + w1 = phase->transform(w1); + w2 = phase->transform(w2); - // Construct both t3 and t4 before transforming so t2 doesn't go dead - // prematurely. - Node *t3 = new (phase->C, 3) RShiftLNode(t2, phase->intcon(N / 2)); - Node *t4 = new (phase->C, 3) AndLNode(t2, phase->longcon(0xFFFFFFFF)); - t3 = phase->transform(t3); - t4 = phase->transform(t4); + // w1 = u0*v1 + w1; + Node* u0v1 = phase->transform(new (phase->C, 3) MulLNode(u0, v1)); + w1 = phase->transform(new (phase->C, 3) AddLNode(u0v1, w1)); - Node *t5 = phase->transform(new (phase->C, 3) AddLNode(t4, lohi_product)); - Node *t6 = phase->transform(new (phase->C, 3) RShiftLNode(t5, phase->intcon(N / 2))); - Node *t7 = phase->transform(new (phase->C, 3) AddLNode(t3, hihi_product)); + // return u1*v1 + w2 + (w1 >> 32); + Node* u1v1 = phase->transform(new (phase->C, 3) MulLNode(u1, v1)); + Node* temp1 = phase->transform(new (phase->C, 3) AddLNode(u1v1, w2)); + Node* temp2 = phase->transform(new (phase->C, 3) RShiftLNode(w1, phase->intcon(N / 2))); - return new (phase->C, 3) AddLNode(t7, t6); + return new (phase->C, 3) AddLNode(temp1, temp2); } @@ -976,7 +1007,7 @@ // Expand mod if( con >= 0 && con < max_jlong && is_power_of_2_long(con+1) ) { - uint k = log2_long(con); // Extract k + uint k = exact_log2_long(con+1); // Extract k // Basic algorithm by David Detlefs. See fastmod_long.java for gory details. // Used to help a popular random number generator which does a long-mod
--- a/src/share/vm/opto/gcm.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/gcm.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -29,6 +29,9 @@ #include "incls/_precompiled.incl" #include "incls/_gcm.cpp.incl" +// To avoid float value underflow +#define MIN_BLOCK_FREQUENCY 1.e-35f + //----------------------------schedule_node_into_block------------------------- // Insert node n into block b. Look for projections of n and make sure they // are in b also. @@ -1380,6 +1383,13 @@ } } +#ifdef ASSERT + for (uint i = 0; i < _num_blocks; i++ ) { + Block *b = _blocks[i]; + assert(b->_freq >= MIN_BLOCK_FREQUENCY, "Register Allocator requiers meaningful block frequency"); + } +#endif + #ifndef PRODUCT if (PrintCFGBlockFreq) { tty->print_cr("CFG Block Frequencies"); @@ -1877,7 +1887,9 @@ float loop_freq = _freq * trip_count(); for (int i = 0; i < _members.length(); i++) { CFGElement* s = _members.at(i); - s->_freq *= loop_freq; + float block_freq = s->_freq * loop_freq; + if (block_freq < MIN_BLOCK_FREQUENCY) block_freq = MIN_BLOCK_FREQUENCY; + s->_freq = block_freq; } CFGLoop* ch = _child; while (ch != NULL) {
--- a/src/share/vm/opto/graphKit.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/graphKit.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -1836,10 +1836,7 @@ (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 - // Get base of card map - assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), - "adjust this code"); - Node *c = makecon(TypeRawPtr::make((address)ct->byte_map_base)); + 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 )); @@ -2945,16 +2942,10 @@ // Now generate allocation code - // With escape analysis, the entire memory state is needed to be able to - // eliminate the allocation. If the allocations cannot be eliminated, this - // will be optimized to the raw slice when the allocation is expanded. - Node *mem; - if (C->do_escape_analysis()) { - mem = reset_memory(); - set_all_memory(mem); - } else { - mem = memory(Compile::AliasIdxRaw); - } + // The entire memory state is needed for slow path of the allocation + // since GC and deoptimization can happened. + Node *mem = reset_memory(); + set_all_memory(mem); // Create new memory state AllocateNode* alloc = new (C, AllocateNode::ParmLimit) @@ -3091,16 +3082,10 @@ // Now generate allocation code - // With escape analysis, the entire memory state is needed to be able to - // eliminate the allocation. If the allocations cannot be eliminated, this - // will be optimized to the raw slice when the allocation is expanded. - Node *mem; - if (C->do_escape_analysis()) { - mem = reset_memory(); - set_all_memory(mem); - } else { - mem = memory(Compile::AliasIdxRaw); - } + // The entire memory state is needed for slow path of the allocation + // since GC and deoptimization can happened. + Node *mem = reset_memory(); + set_all_memory(mem); // Create new memory state // Create the AllocateArrayNode and its result projections AllocateArrayNode* alloc @@ -3233,12 +3218,11 @@ // Now some of the values - Node* marking = __ load(no_ctrl, marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw); - Node* index = __ load(no_ctrl, index_adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw); - Node* buffer = __ load(no_ctrl, buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw); + Node* marking = __ load(__ ctrl(), marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw); // if (!marking) __ if_then(marking, BoolTest::ne, zero); { + Node* index = __ load(__ ctrl(), index_adr, TypeInt::INT, T_INT, Compile::AliasIdxRaw); const Type* t1 = adr->bottom_type(); const Type* t2 = val->bottom_type(); @@ -3246,6 +3230,7 @@ Node* orig = __ load(no_ctrl, adr, val_type, bt, alias_idx); // if (orig != NULL) __ if_then(orig, BoolTest::ne, null()); { + Node* buffer = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw); // load original value // alias_idx correct?? @@ -3365,14 +3350,6 @@ const TypeFunc *tf = OptoRuntime::g1_wb_post_Type(); - // Get the address of the card table - CardTableModRefBS* ct = - (CardTableModRefBS*)(Universe::heap()->barrier_set()); - Node *card_table = __ makecon(TypeRawPtr::make((address)ct->byte_map_base)); - // Get base of card map - assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); - - // Offsets into the thread const int index_offset = in_bytes(JavaThread::dirty_card_queue_offset() + PtrQueue::byte_offset_of_index()); @@ -3402,7 +3379,7 @@ Node* card_offset = __ URShiftX( cast, __ ConI(CardTableModRefBS::card_shift) ); // Combine card table base and card offset - Node *card_adr = __ AddP(no_base, card_table, 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?
--- a/src/share/vm/opto/graphKit.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/graphKit.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -83,6 +83,18 @@ Node* zerocon(BasicType bt) const { return _gvn.zerocon(bt); } // (See also macro MakeConX in type.hpp, which uses intcon or longcon.) + // Helper for byte_map_base + Node* byte_map_base_node() { + // Get base of card map + CardTableModRefBS* ct = (CardTableModRefBS*)(Universe::heap()->barrier_set()); + assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust users of this code"); + if (ct->byte_map_base != NULL) { + return makecon(TypeRawPtr::make((address)ct->byte_map_base)); + } else { + return null(); + } + } + jint find_int_con(Node* n, jint value_if_unknown) { return _gvn.find_int_con(n, value_if_unknown); }
--- a/src/share/vm/opto/ifg.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/ifg.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -471,12 +471,28 @@ // for the "collect_gc_info" phase later. IndexSet liveout(_live->live(b)); uint last_inst = b->end_idx(); - // Compute last phi index - uint last_phi; - for( last_phi = 1; last_phi < last_inst; last_phi++ ) - if( !b->_nodes[last_phi]->is_Phi() ) + // Compute first nonphi node index + uint first_inst; + for( first_inst = 1; first_inst < last_inst; first_inst++ ) + if( !b->_nodes[first_inst]->is_Phi() ) break; + // Spills could be inserted before CreateEx node which should be + // first instruction in block after Phis. Move CreateEx up. + for( uint insidx = first_inst; insidx < last_inst; insidx++ ) { + Node *ex = b->_nodes[insidx]; + if( ex->is_SpillCopy() ) continue; + if( insidx > first_inst && ex->is_Mach() && + ex->as_Mach()->ideal_Opcode() == Op_CreateEx ) { + // If the CreateEx isn't above all the MachSpillCopies + // then move it to the top. + b->_nodes.remove(insidx); + b->_nodes.insert(first_inst, ex); + } + // Stop once a CreateEx or any other node is found + break; + } + // Reset block's register pressure values for each ifg construction uint pressure[2], hrp_index[2]; pressure[0] = pressure[1] = 0; @@ -485,7 +501,7 @@ // Liveout things are presumed live for the whole block. We accumulate // 'area' accordingly. If they get killed in the block, we'll subtract // the unused part of the block from the area. - int inst_count = last_inst - last_phi; + int inst_count = last_inst - first_inst; double cost = (inst_count <= 0) ? 0.0 : b->_freq * double(inst_count); assert(!(cost < 0.0), "negative spill cost" ); IndexSetIterator elements(&liveout);
--- a/src/share/vm/opto/lcm.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/lcm.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -107,7 +107,7 @@ was_store = false; switch( mach->ideal_Opcode() ) { case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadD: case Op_LoadF: case Op_LoadI:
--- a/src/share/vm/opto/live.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/live.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -271,9 +271,9 @@ //------------------------------verify_base_ptrs------------------------------- // Verify that base pointers and derived pointers are still sane. -// Basically, if a derived pointer is live at a safepoint, then its -// base pointer must be live also. void PhaseChaitin::verify_base_ptrs( ResourceArea *a ) const { +#ifdef ASSERT + Unique_Node_List worklist(a); for( uint i = 0; i < _cfg._num_blocks; i++ ) { Block *b = _cfg._blocks[i]; for( uint j = b->end_idx() + 1; j > 1; j-- ) { @@ -287,28 +287,81 @@ // Now scan for a live derived pointer if (jvms->oopoff() < sfpt->req()) { // Check each derived/base pair - for (uint idx = jvms->oopoff(); idx < sfpt->req(); idx += 2) { + for (uint idx = jvms->oopoff(); idx < sfpt->req(); idx++) { Node *check = sfpt->in(idx); - uint j = 0; + bool is_derived = ((idx - jvms->oopoff()) & 1) == 0; // search upwards through spills and spill phis for AddP - while(true) { - if( !check ) break; - int idx = check->is_Copy(); - if( idx ) { - check = check->in(idx); - } else if( check->is_Phi() && check->_idx >= _oldphi ) { - check = check->in(1); - } else - break; - j++; - assert(j < 100000,"Derived pointer checking in infinite loop"); + worklist.clear(); + worklist.push(check); + uint k = 0; + while( k < worklist.size() ) { + check = worklist.at(k); + assert(check,"Bad base or derived pointer"); + // See PhaseChaitin::find_base_for_derived() for all cases. + int isc = check->is_Copy(); + if( isc ) { + worklist.push(check->in(isc)); + } else if( check->is_Phi() ) { + for (uint m = 1; m < check->req(); m++) + worklist.push(check->in(m)); + } else if( check->is_Con() ) { + if (is_derived) { + // Derived is NULL+offset + assert(!is_derived || check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad derived pointer"); + } else { + assert(check->bottom_type()->is_ptr()->_offset == 0,"Bad base pointer"); + // Base either ConP(NULL) or loadConP + if (check->is_Mach()) { + assert(check->as_Mach()->ideal_Opcode() == Op_ConP,"Bad base pointer"); + } else { + assert(check->Opcode() == Op_ConP && + check->bottom_type()->is_ptr()->ptr() == TypePtr::Null,"Bad base pointer"); + } + } + } else if( check->bottom_type()->is_ptr()->_offset == 0 ) { + if(check->is_Proj() || check->is_Mach() && + (check->as_Mach()->ideal_Opcode() == Op_CreateEx || + check->as_Mach()->ideal_Opcode() == Op_ThreadLocal || + check->as_Mach()->ideal_Opcode() == Op_CMoveP || + check->as_Mach()->ideal_Opcode() == Op_CheckCastPP || +#ifdef _LP64 + UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_CastPP || + UseCompressedOops && check->as_Mach()->ideal_Opcode() == Op_DecodeN || +#endif + check->as_Mach()->ideal_Opcode() == Op_LoadP || + check->as_Mach()->ideal_Opcode() == Op_LoadKlass)) { + // Valid nodes + } else { + check->dump(); + assert(false,"Bad base or derived pointer"); + } + } else { + assert(is_derived,"Bad base pointer"); + assert(check->is_Mach() && check->as_Mach()->ideal_Opcode() == Op_AddP,"Bad derived pointer"); + } + k++; + assert(k < 100000,"Derived pointer checking in infinite loop"); } // End while - assert(check->is_Mach() && check->as_Mach()->ideal_Opcode() == Op_AddP,"Bad derived pointer") } } // End of check for derived pointers } // End of Kcheck for debug info } // End of if found a safepoint } // End of forall instructions in block } // End of forall blocks +#endif } + +//------------------------------verify------------------------------------- +// Verify that graphs and base pointers are still sane. +void PhaseChaitin::verify( ResourceArea *a, bool verify_ifg ) const { +#ifdef ASSERT + if( VerifyOpto || VerifyRegisterAllocator ) { + _cfg.verify(); + verify_base_ptrs(a); + if(verify_ifg) + _ifg->verify(this); + } #endif +} + +#endif
--- a/src/share/vm/opto/loopnode.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/loopnode.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -2654,7 +2654,7 @@ case Op_ModF: case Op_ModD: case Op_LoadB: // Same with Loads; they can sink - case Op_LoadC: // during loop optimizations. + case Op_LoadUS: // during loop optimizations. case Op_LoadD: case Op_LoadF: case Op_LoadI:
--- a/src/share/vm/opto/macro.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/macro.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -952,13 +952,6 @@ Node* klass_node = alloc->in(AllocateNode::KlassNode); Node* initial_slow_test = alloc->in(AllocateNode::InitialTest); - // With escape analysis, the entire memory state was needed to be able to - // eliminate the allocation. Since the allocations cannot be eliminated, - // optimize it to the raw slice. - if (mem->is_MergeMem()) { - mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw); - } - assert(ctrl != NULL, "must have control"); // We need a Region and corresponding Phi's to merge the slow-path and fast-path results. // they will not be used if "always_slow" is set @@ -1016,6 +1009,11 @@ Node *slow_mem = mem; // save the current memory state for slow path // generate the fast allocation code unless we know that the initial test will always go slow if (!always_slow) { + // Fast path modifies only raw memory. + if (mem->is_MergeMem()) { + mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw); + } + Node* eden_top_adr; Node* eden_end_adr; @@ -1239,8 +1237,6 @@ } } - mem = result_phi_rawmem; - // An allocate node has separate i_o projections for the uses on the control and i_o paths // Replace uses of the control i_o projection with result_phi_i_o (unless we are only generating a slow call) if (_ioproj_fallthrough == NULL) {
--- a/src/share/vm/opto/matcher.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/matcher.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -1824,7 +1824,7 @@ mem_op = true; break; case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadD: case Op_LoadF: case Op_LoadI:
--- a/src/share/vm/opto/memnode.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/memnode.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -779,14 +779,14 @@ "use LoadRangeNode instead"); switch (bt) { case T_BOOLEAN: - case T_BYTE: return new (C, 3) LoadBNode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_INT: return new (C, 3) LoadINode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_CHAR: return new (C, 3) LoadCNode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_SHORT: return new (C, 3) LoadSNode(ctl, mem, adr, adr_type, rt->is_int() ); - case T_LONG: return new (C, 3) LoadLNode(ctl, mem, adr, adr_type, rt->is_long() ); - case T_FLOAT: return new (C, 3) LoadFNode(ctl, mem, adr, adr_type, rt ); - case T_DOUBLE: return new (C, 3) LoadDNode(ctl, mem, adr, adr_type, rt ); - case T_ADDRESS: return new (C, 3) LoadPNode(ctl, mem, adr, adr_type, rt->is_ptr() ); + case T_BYTE: return new (C, 3) LoadBNode (ctl, mem, adr, adr_type, rt->is_int() ); + case T_INT: return new (C, 3) LoadINode (ctl, mem, adr, adr_type, rt->is_int() ); + case T_CHAR: return new (C, 3) LoadUSNode(ctl, mem, adr, adr_type, rt->is_int() ); + case T_SHORT: return new (C, 3) LoadSNode (ctl, mem, adr, adr_type, rt->is_int() ); + case T_LONG: return new (C, 3) LoadLNode (ctl, mem, adr, adr_type, rt->is_long() ); + case T_FLOAT: return new (C, 3) LoadFNode (ctl, mem, adr, adr_type, rt ); + case T_DOUBLE: return new (C, 3) LoadDNode (ctl, mem, adr, adr_type, rt ); + case T_ADDRESS: return new (C, 3) LoadPNode (ctl, mem, adr, adr_type, rt->is_ptr() ); case T_OBJECT: #ifdef _LP64 if (adr->bottom_type()->is_ptr_to_narrowoop()) { @@ -1076,13 +1076,14 @@ // of the original value. Node* mem_phi = in(Memory); Node* offset = in(Address)->in(AddPNode::Offset); + Node* region = base->in(0); Node* in1 = clone(); Node* in1_addr = in1->in(Address)->clone(); in1_addr->set_req(AddPNode::Base, base->in(allocation_index)); in1_addr->set_req(AddPNode::Address, base->in(allocation_index)); in1_addr->set_req(AddPNode::Offset, offset); - in1->set_req(0, base->in(allocation_index)); + in1->set_req(0, region->in(allocation_index)); in1->set_req(Address, in1_addr); in1->set_req(Memory, mem_phi->in(allocation_index)); @@ -1091,7 +1092,7 @@ in2_addr->set_req(AddPNode::Base, base->in(load_index)); in2_addr->set_req(AddPNode::Address, base->in(load_index)); in2_addr->set_req(AddPNode::Offset, offset); - in2->set_req(0, base->in(load_index)); + in2->set_req(0, region->in(load_index)); in2->set_req(Address, in2_addr); in2->set_req(Memory, mem_phi->in(load_index)); @@ -1100,7 +1101,7 @@ in2_addr = phase->transform(in2_addr); in2 = phase->transform(in2); - PhiNode* result = PhiNode::make_blank(base->in(0), this); + PhiNode* result = PhiNode::make_blank(region, this); result->set_req(allocation_index, in1); result->set_req(load_index, in2); return result; @@ -1303,6 +1304,7 @@ Node* base = AddPNode::Ideal_base_and_offset(address, phase, ignore); if (base != NULL && phase->type(base)->higher_equal(TypePtr::NOTNULL) + && phase->C->get_alias_index(phase->type(address)->is_ptr()) != Compile::AliasIdxRaw && all_controls_dominate(base, phase->C->start())) { // A method-invariant, non-null address (constant or 'this' argument). set_req(MemNode::Control, NULL); @@ -1356,7 +1358,7 @@ // Steps (a), (b): Walk past independent stores to find an exact match. if (prev_mem != NULL && prev_mem != in(MemNode::Memory)) { // (c) See if we can fold up on the spot, but don't fold up here. - // Fold-up might require truncation (for LoadB/LoadS/LoadC) or + // Fold-up might require truncation (for LoadB/LoadS/LoadUS) or // just return a prior value, which is done by Identity calls. if (can_see_stored_value(prev_mem, phase)) { // Make ready for step (d): @@ -1605,14 +1607,14 @@ return LoadNode::Ideal(phase, can_reshape); } -//--------------------------LoadCNode::Ideal-------------------------------------- +//--------------------------LoadUSNode::Ideal------------------------------------- // // If the previous store is to the same address as this load, // and the value stored was larger than a char, replace this load // with the value stored truncated to a char. If no truncation is // needed, the replacement is done in LoadNode::Identity(). // -Node *LoadCNode::Ideal(PhaseGVN *phase, bool can_reshape) { +Node *LoadUSNode::Ideal(PhaseGVN *phase, bool can_reshape) { Node* mem = in(MemNode::Memory); Node* value = can_see_stored_value(mem,phase); if( value && !phase->type(value)->higher_equal( _type ) )
--- a/src/share/vm/opto/memnode.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/memnode.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -207,11 +207,11 @@ virtual BasicType memory_type() const { return T_BYTE; } }; -//------------------------------LoadCNode-------------------------------------- -// Load a char (16bits unsigned) from memory -class LoadCNode : public LoadNode { +//------------------------------LoadUSNode------------------------------------- +// Load an unsigned short/char (16bits unsigned) from memory +class LoadUSNode : public LoadNode { public: - LoadCNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::CHAR ) + LoadUSNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::CHAR ) : LoadNode(c,mem,adr,at,ti) {} virtual int Opcode() const; virtual uint ideal_reg() const { return Op_RegI; }
--- a/src/share/vm/opto/mulnode.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/mulnode.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 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 @@ -442,16 +442,17 @@ return load; } uint lop = load->Opcode(); - if( lop == Op_LoadC && + if( lop == Op_LoadUS && con == 0x0000FFFF ) // Already zero-extended return load; // Masking off the high bits of a unsigned-shift-right is not // needed either. if( lop == Op_URShiftI ) { const TypeInt *t12 = phase->type( load->in(2) )->isa_int(); - if( t12 && t12->is_con() ) { - int shift_con = t12->get_con(); - int mask = max_juint >> shift_con; + if( t12 && t12->is_con() ) { // Shift is by a constant + int shift = t12->get_con(); + shift &= BitsPerJavaInteger - 1; // semantics of Java shifts + int mask = max_juint >> shift; if( (mask&con) == mask ) // If AND is useless, skip it return load; } @@ -470,19 +471,19 @@ uint lop = load->Opcode(); // Masking bits off of a Character? Hi bits are already zero. - if( lop == Op_LoadC && + if( lop == Op_LoadUS && (mask & 0xFFFF0000) ) // Can we make a smaller mask? return new (phase->C, 3) AndINode(load,phase->intcon(mask&0xFFFF)); // Masking bits off of a Short? Loading a Character does some masking if( lop == Op_LoadS && (mask & 0xFFFF0000) == 0 ) { - Node *ldc = new (phase->C, 3) LoadCNode(load->in(MemNode::Control), + Node *ldus = new (phase->C, 3) LoadUSNode(load->in(MemNode::Control), load->in(MemNode::Memory), load->in(MemNode::Address), load->adr_type()); - ldc = phase->transform(ldc); - return new (phase->C, 3) AndINode(ldc,phase->intcon(mask&0xFFFF)); + ldus = phase->transform(ldus); + return new (phase->C, 3) AndINode(ldus, phase->intcon(mask&0xFFFF)); } // Masking sign bits off of a Byte? Let the matcher use an unsigned load @@ -579,9 +580,10 @@ // needed either. if( lop == Op_URShiftL ) { const TypeInt *t12 = phase->type( usr->in(2) )->isa_int(); - if( t12 && t12->is_con() ) { - int shift_con = t12->get_con(); - jlong mask = max_julong >> shift_con; + if( t12 && t12->is_con() ) { // Shift is by a constant + int shift = t12->get_con(); + shift &= BitsPerJavaLong - 1; // semantics of Java shifts + jlong mask = max_julong >> shift; if( (mask&con) == mask ) // If AND is useless, skip it return usr; } @@ -605,8 +607,8 @@ const TypeInt *t12 = phase->type(rsh->in(2))->isa_int(); if( t12 && t12->is_con() ) { // Shift is by a constant int shift = t12->get_con(); - shift &= (BitsPerJavaInteger*2)-1; // semantics of Java shifts - const jlong sign_bits_mask = ~(((jlong)CONST64(1) << (jlong)(BitsPerJavaInteger*2 - shift)) -1); + shift &= BitsPerJavaLong - 1; // semantics of Java shifts + const jlong sign_bits_mask = ~(((jlong)CONST64(1) << (jlong)(BitsPerJavaLong - shift)) -1); // If the AND'ing of the 2 masks has no bits, then only original shifted // bits survive. NO sign-extension bits survive the maskings. if( (sign_bits_mask & mask) == 0 ) { @@ -786,7 +788,7 @@ // Check for ((x & ((CONST64(1)<<(64-c0))-1)) << c0) which ANDs off high bits // before shifting them away. - const jlong bits_mask = ((jlong)CONST64(1) << (jlong)(BitsPerJavaInteger*2 - con)) - CONST64(1); + const jlong bits_mask = ((jlong)CONST64(1) << (jlong)(BitsPerJavaLong - con)) - CONST64(1); if( add1_op == Op_AndL && phase->type(add1->in(2)) == TypeLong::make( bits_mask ) ) return new (phase->C, 3) LShiftLNode( add1->in(1), in(2) ); @@ -820,7 +822,7 @@ return TypeLong::LONG; uint shift = r2->get_con(); - shift &= (BitsPerJavaInteger*2)-1; // semantics of Java shifts + shift &= BitsPerJavaLong - 1; // semantics of Java shifts // Shift by a multiple of 64 does nothing: if (shift == 0) return t1; @@ -913,7 +915,7 @@ set_req(2, phase->intcon(0)); return this; } - else if( ld->Opcode() == Op_LoadC ) + else if( ld->Opcode() == Op_LoadUS ) // Replace zero-extension-load with sign-extension-load return new (phase->C, 3) LoadSNode( ld->in(MemNode::Control), ld->in(MemNode::Memory), @@ -1235,7 +1237,7 @@ if ( con == 0 ) return NULL; // let Identity() handle a 0 shift count // note: mask computation below does not work for 0 shift count // We'll be wanting the right-shift amount as a mask of that many bits - const jlong mask = (((jlong)CONST64(1) << (jlong)(BitsPerJavaInteger*2 - con)) -1); + const jlong mask = (((jlong)CONST64(1) << (jlong)(BitsPerJavaLong - con)) -1); // Check for ((x << z) + Y) >>> z. Replace with x + con>>>z // The idiom for rounding to a power of 2 is "(Q+(2^z-1)) >>> z". @@ -1302,7 +1304,7 @@ if (r2->is_con()) { uint shift = r2->get_con(); - shift &= (2*BitsPerJavaInteger)-1; // semantics of Java shifts + shift &= BitsPerJavaLong - 1; // semantics of Java shifts // Shift by a multiple of 64 does nothing: if (shift == 0) return t1; // Calculate reasonably aggressive bounds for the result. @@ -1325,7 +1327,7 @@ const TypeLong* tl = TypeLong::make(lo, hi, MAX2(r1->_widen,r2->_widen)); #ifdef ASSERT // Make sure we get the sign-capture idiom correct. - if (shift == (2*BitsPerJavaInteger)-1) { + if (shift == BitsPerJavaLong - 1) { if (r1->_lo >= 0) assert(tl == TypeLong::ZERO, ">>>63 of + is 0"); if (r1->_hi < 0) assert(tl == TypeLong::ONE, ">>>63 of - is +1"); }
--- a/src/share/vm/opto/superword.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/superword.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -1444,7 +1444,7 @@ // (Start, end] half-open range defining which operands are vector void SuperWord::vector_opd_range(Node* n, uint* start, uint* end) { switch (n->Opcode()) { - case Op_LoadB: case Op_LoadC: + case Op_LoadB: case Op_LoadUS: case Op_LoadI: case Op_LoadL: case Op_LoadF: case Op_LoadD: case Op_LoadP:
--- a/src/share/vm/opto/type.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/type.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -2471,6 +2471,8 @@ const Type* ft = join(kills); const TypeInstPtr* ftip = ft->isa_instptr(); const TypeInstPtr* ktip = kills->isa_instptr(); + const TypeKlassPtr* ftkp = ft->isa_klassptr(); + const TypeKlassPtr* ktkp = kills->isa_klassptr(); if (ft->empty()) { // Check for evil case of 'this' being a class and 'kills' expecting an @@ -2484,6 +2486,8 @@ // uplift the type. if (!empty() && ktip != NULL && ktip->is_loaded() && ktip->klass()->is_interface()) return kills; // Uplift to interface + if (!empty() && ktkp != NULL && ktkp->klass()->is_loaded() && ktkp->klass()->is_interface()) + return kills; // Uplift to interface return Type::TOP; // Canonical empty value } @@ -2499,6 +2503,12 @@ // Happens in a CTW of rt.jar, 320-341, no extra flags return ktip->cast_to_ptr_type(ftip->ptr()); } + if (ftkp != NULL && ktkp != NULL && + ftkp->is_loaded() && ftkp->klass()->is_interface() && + ktkp->is_loaded() && !ktkp->klass()->is_interface()) { + // Happens in a CTW of rt.jar, 320-341, no extra flags + return ktkp->cast_to_ptr_type(ftkp->ptr()); + } return ft; } @@ -3657,7 +3667,7 @@ //------------------------------cast_to_ptr_type------------------------------- const Type *TypeKlassPtr::cast_to_ptr_type(PTR ptr) const { - assert(_base == OopPtr, "subclass must override cast_to_ptr_type"); + assert(_base == KlassPtr, "subclass must override cast_to_ptr_type"); if( ptr == _ptr ) return this; return make(ptr, _klass, _offset); }
--- a/src/share/vm/opto/type.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/type.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -882,6 +882,8 @@ public: ciSymbol* name() const { return _klass->name(); } + bool is_loaded() const { return _klass->is_loaded(); } + // ptr to klass 'k' static const TypeKlassPtr *make( ciKlass* k ) { return make( TypePtr::Constant, k, 0); } // ptr to klass 'k' with offset
--- a/src/share/vm/opto/vectornode.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/opto/vectornode.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -239,7 +239,7 @@ return Op_XorV; case Op_LoadB: - case Op_LoadC: + case Op_LoadUS: case Op_LoadS: case Op_LoadI: case Op_LoadL: @@ -269,7 +269,7 @@ case 16: return Op_Load16B; } break; - case Op_LoadC: + case Op_LoadUS: switch (vlen) { case 2: return Op_Load2C; case 4: return Op_Load4C;
--- a/src/share/vm/runtime/arguments.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/runtime/arguments.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -2489,7 +2489,7 @@ vm_args.version = JNI_VERSION_1_2; vm_args.options = options; vm_args.nOptions = i; - vm_args.ignoreUnrecognized = false; + vm_args.ignoreUnrecognized = IgnoreUnrecognizedVMOptions; if (PrintVMOptions) { const char* tail; @@ -2536,13 +2536,12 @@ // If flag "-XX:Flags=flags-file" is used it will be the first option to be processed. bool settings_file_specified = false; + const char* flags_file; int index; for (index = 0; index < args->nOptions; index++) { const JavaVMOption *option = args->options + index; if (match_option(option, "-XX:Flags=", &tail)) { - if (!process_settings_file(tail, true, args->ignoreUnrecognized)) { - return JNI_EINVAL; - } + flags_file = tail; settings_file_specified = true; } if (match_option(option, "-XX:+PrintVMOptions", &tail)) { @@ -2551,6 +2550,24 @@ if (match_option(option, "-XX:-PrintVMOptions", &tail)) { PrintVMOptions = false; } + if (match_option(option, "-XX:+IgnoreUnrecognizedVMOptions", &tail)) { + IgnoreUnrecognizedVMOptions = true; + } + if (match_option(option, "-XX:-IgnoreUnrecognizedVMOptions", &tail)) { + IgnoreUnrecognizedVMOptions = false; + } + } + + if (IgnoreUnrecognizedVMOptions) { + // uncast const to modify the flag args->ignoreUnrecognized + *(jboolean*)(&args->ignoreUnrecognized) = true; + } + + // Parse specified settings file + if (settings_file_specified) { + if (!process_settings_file(flags_file, true, args->ignoreUnrecognized)) { + return JNI_EINVAL; + } } // Parse default .hotspotrc settings file
--- a/src/share/vm/runtime/globals.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/runtime/globals.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -1426,10 +1426,10 @@ develop(bool, CMSOverflowEarlyRestoration, false, \ "Whether preserved marks should be restored early") \ \ - product(uintx, CMSMarkStackSize, 32*K, \ + product(uintx, CMSMarkStackSize, NOT_LP64(32*K) LP64_ONLY(4*M), \ "Size of CMS marking stack") \ \ - product(uintx, CMSMarkStackSizeMax, 4*M, \ + product(uintx, CMSMarkStackSizeMax, NOT_LP64(4*M) LP64_ONLY(512*M), \ "Max size of CMS marking stack") \ \ notproduct(bool, CMSMarkStackOverflowALot, false, \ @@ -1655,6 +1655,13 @@ develop(uintx, WorkStealingYieldsBeforeSleep, 1000, \ "Number of yields before a sleep is done during workstealing") \ \ + develop(uintx, WorkStealingHardSpins, 4096, \ + "Number of iterations in a spin loop between checks on " \ + "time out of hard spin") \ + \ + develop(uintx, WorkStealingSpinToYieldRatio, 10, \ + "Ratio of hard spins to calls to yield") \ + \ product(uintx, PreserveMarkStackSize, 1024, \ "Size for stack used in promotion failure handling") \ \ @@ -2187,6 +2194,9 @@ product(bool, PrintVMOptions, trueInDebug, \ "print VM flag settings") \ \ + product(bool, IgnoreUnrecognizedVMOptions, false, \ + "Ignore unrecognized VM options") \ + \ diagnostic(bool, SerializeVMOutput, true, \ "Use a mutex to serialize output to tty and hotspot.log") \ \
--- a/src/share/vm/runtime/os.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/runtime/os.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -74,13 +74,11 @@ const int milliseconds_after_second = milliseconds_since_19700101 % milliseconds_per_microsecond; // Convert the time value to a tm and timezone variable - const struct tm *time_struct_temp = localtime(&seconds_since_19700101); - if (time_struct_temp == NULL) { - assert(false, "Failed localtime"); + struct tm time_struct; + if (localtime_pd(&seconds_since_19700101, &time_struct) == NULL) { + assert(false, "Failed localtime_pd"); return NULL; } - // Save the results of localtime - const struct tm time_struct = *time_struct_temp; const time_t zone = timezone; // If daylight savings time is in effect, @@ -93,10 +91,10 @@ UTC_to_local = UTC_to_local - seconds_per_hour; } // Compute the time zone offset. - // localtime(3C) sets timezone to the difference (in seconds) + // localtime_pd() sets timezone to the difference (in seconds) // between UTC and and local time. // ISO 8601 says we need the difference between local time and UTC, - // we change the sign of the localtime(3C) result. + // we change the sign of the localtime_pd() result. const time_t local_to_UTC = -(UTC_to_local); // Then we have to figure out if if we are ahead (+) or behind (-) UTC. char sign_local_to_UTC = '+';
--- a/src/share/vm/runtime/os.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/runtime/os.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -120,7 +120,8 @@ // Return current local time in a string (YYYY-MM-DD HH:MM:SS). // It is MT safe, but not async-safe, as reading time zone // information may require a lock on some platforms. - static char* local_time_string(char *buf, size_t buflen); + static char* local_time_string(char *buf, size_t buflen); + static struct tm* localtime_pd (const time_t* clock, struct tm* res); // Fill in buffer with current local time as an ISO-8601 string. // E.g., YYYY-MM-DDThh:mm:ss.mmm+zzzz. // Returns buffer, or NULL if it failed.
--- a/src/share/vm/utilities/globalDefinitions.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/utilities/globalDefinitions.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -1,5 +1,5 @@ /* - * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved. + * Copyright 1997-2009 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 @@ -74,6 +74,7 @@ extern int BitsPerHeapOop; const int BitsPerJavaInteger = 32; +const int BitsPerJavaLong = 64; const int BitsPerSize_t = size_tSize * BitsPerByte; // Size of a char[] needed to represent a jint as a string in decimal. @@ -906,6 +907,14 @@ return log2_intptr(x); } +//* the argument must be exactly a power of 2 +inline int exact_log2_long(jlong x) { + #ifdef ASSERT + if (!is_power_of_2_long(x)) basic_fatal("x must be a power of 2"); + #endif + return log2_long(x); +} + // returns integer round-up to the nearest multiple of s (s must be a power of two) inline intptr_t round_to(intptr_t x, uintx s) {
--- a/src/share/vm/utilities/taskqueue.cpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/utilities/taskqueue.cpp Wed Feb 25 23:16:22 2009 -0800 @@ -25,6 +25,12 @@ # include "incls/_precompiled.incl" # include "incls/_taskqueue.cpp.incl" +#ifdef TRACESPINNING +uint ParallelTaskTerminator::_total_yields = 0; +uint ParallelTaskTerminator::_total_spins = 0; +uint ParallelTaskTerminator::_total_peeks = 0; +#endif + bool TaskQueueSuper::peek() { return _bottom != _age.top(); } @@ -69,15 +75,62 @@ ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) { Atomic::inc(&_offered_termination); - juint yield_count = 0; + uint yield_count = 0; + // Number of hard spin loops done since last yield + uint hard_spin_count = 0; + // Number of iterations in the hard spin loop. + uint hard_spin_limit = WorkStealingHardSpins; + + // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done. + // If it is greater than 0, then start with a small number + // of spins and increase number with each turn at spinning until + // the count of hard spins exceeds WorkStealingSpinToYieldRatio. + // Then do a yield() call and start spinning afresh. + if (WorkStealingSpinToYieldRatio > 0) { + hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio; + hard_spin_limit = MAX2(hard_spin_limit, 1U); + } + // Remember the initial spin limit. + uint hard_spin_start = hard_spin_limit; + + // Loop waiting for all threads to offer termination or + // more work. while (true) { + // Are all threads offering termination? if (_offered_termination == _n_threads) { - //inner_termination_loop(); return true; } else { + // Look for more work. + // Periodically sleep() instead of yield() to give threads + // waiting on the cores the chance to grab this code if (yield_count <= WorkStealingYieldsBeforeSleep) { + // Do a yield or hardspin. For purposes of deciding whether + // to sleep, count this as a yield. yield_count++; - yield(); + + // Periodically call yield() instead spinning + // After WorkStealingSpinToYieldRatio spins, do a yield() call + // and reset the counts and starting limit. + if (hard_spin_count > WorkStealingSpinToYieldRatio) { + yield(); + hard_spin_count = 0; + hard_spin_limit = hard_spin_start; +#ifdef TRACESPINNING + _total_yields++; +#endif + } else { + // Hard spin this time + // Increase the hard spinning period but only up to a limit. + hard_spin_limit = MIN2(2*hard_spin_limit, + (uint) WorkStealingHardSpins); + for (uint j = 0; j < hard_spin_limit; j++) { + SpinPause(); + } + hard_spin_count++; +#ifdef TRACESPINNING + _total_spins++; +#endif + } } else { if (PrintGCDetails && Verbose) { gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() " @@ -92,6 +145,9 @@ sleep(WorkStealingSleepMillis); } +#ifdef TRACESPINNING + _total_peeks++; +#endif if (peek_in_queue_set() || (terminator != NULL && terminator->should_exit_termination())) { Atomic::dec(&_offered_termination); @@ -101,6 +157,16 @@ } } +#ifdef TRACESPINNING +void ParallelTaskTerminator::print_termination_counts() { + gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld " + "Total spins: %lld Total peeks: %lld", + total_yields(), + total_spins(), + total_peeks()); +} +#endif + void ParallelTaskTerminator::reset_for_reuse() { if (_offered_termination != 0) { assert(_offered_termination == _n_threads,
--- a/src/share/vm/utilities/taskqueue.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/utilities/taskqueue.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -22,67 +22,76 @@ * */ +#ifdef LP64 +typedef juint TAG_TYPE; +// for a taskqueue size of 4M +#define LOG_TASKQ_SIZE 22 +#else +typedef jushort TAG_TYPE; +// for a taskqueue size of 16K +#define LOG_TASKQ_SIZE 14 +#endif + class TaskQueueSuper: public CHeapObj { protected: // The first free element after the last one pushed (mod _n). - // (For now we'll assume only 32-bit CAS). - volatile juint _bottom; + volatile uint _bottom; // log2 of the size of the queue. enum SomeProtectedConstants { - Log_n = 14 + Log_n = LOG_TASKQ_SIZE }; +#undef LOG_TASKQ_SIZE // Size of the queue. - juint n() { return (1 << Log_n); } + uint n() { return (1 << Log_n); } // For computing "x mod n" efficiently. - juint n_mod_mask() { return n() - 1; } + uint n_mod_mask() { return n() - 1; } struct Age { - jushort _top; - jushort _tag; + TAG_TYPE _top; + TAG_TYPE _tag; - jushort tag() const { return _tag; } - jushort top() const { return _top; } + TAG_TYPE tag() const { return _tag; } + TAG_TYPE top() const { return _top; } Age() { _tag = 0; _top = 0; } friend bool operator ==(const Age& a1, const Age& a2) { return a1.tag() == a2.tag() && a1.top() == a2.top(); } - }; Age _age; // These make sure we do single atomic reads and writes. Age get_age() { - jint res = *(volatile jint*)(&_age); + uint res = *(volatile uint*)(&_age); return *(Age*)(&res); } void set_age(Age a) { - *(volatile jint*)(&_age) = *(int*)(&a); + *(volatile uint*)(&_age) = *(uint*)(&a); } - jushort get_top() { + TAG_TYPE get_top() { return get_age().top(); } // These both operate mod _n. - juint increment_index(juint ind) { + uint increment_index(uint ind) { return (ind + 1) & n_mod_mask(); } - juint decrement_index(juint ind) { + uint decrement_index(uint ind) { return (ind - 1) & n_mod_mask(); } // Returns a number in the range [0.._n). If the result is "n-1", it // should be interpreted as 0. - juint dirty_size(juint bot, juint top) { - return ((jint)bot - (jint)top) & n_mod_mask(); + uint dirty_size(uint bot, uint top) { + return ((int)bot - (int)top) & n_mod_mask(); } // Returns the size corresponding to the given "bot" and "top". - juint size(juint bot, juint top) { - juint sz = dirty_size(bot, top); + uint size(uint bot, uint top) { + uint sz = dirty_size(bot, top); // Has the queue "wrapped", so that bottom is less than top? // There's a complicated special case here. A pair of threads could // perform pop_local and pop_global operations concurrently, starting @@ -94,7 +103,7 @@ // owner performs pop_local's, and several concurrent threads // attempting to perform the pop_global will all perform the same CAS, // and only one can succeed. Any stealing thread that reads after - // either the increment or decrement will seen an empty queue, and will + // either the increment or decrement will see an empty queue, and will // not join the competitors. The "sz == -1 || sz == _n-1" state will // not be modified by concurrent queues, so the owner thread can reset // the state to _bottom == top so subsequent pushes will be performed @@ -112,11 +121,11 @@ // Return an estimate of the number of elements in the queue. // The "careful" version admits the possibility of pop_local/pop_global // races. - juint size() { + uint size() { return size(_bottom, get_top()); } - juint dirty_size() { + uint dirty_size() { return dirty_size(_bottom, get_top()); } @@ -127,15 +136,15 @@ // Maximum number of elements allowed in the queue. This is two less // than the actual queue size, for somewhat complicated reasons. - juint max_elems() { return n() - 2; } + uint max_elems() { return n() - 2; } }; template<class E> class GenericTaskQueue: public TaskQueueSuper { private: // Slow paths for push, pop_local. (pop_global has no fast path.) - bool push_slow(E t, juint dirty_n_elems); - bool pop_local_slow(juint localBot, Age oldAge); + bool push_slow(E t, uint dirty_n_elems); + bool pop_local_slow(uint localBot, Age oldAge); public: // Initializes the queue to empty. @@ -170,7 +179,7 @@ template<class E> GenericTaskQueue<E>::GenericTaskQueue():TaskQueueSuper() { - assert(sizeof(Age) == sizeof(jint), "Depends on this."); + assert(sizeof(Age) == sizeof(int), "Depends on this."); } template<class E> @@ -182,9 +191,9 @@ template<class E> void GenericTaskQueue<E>::oops_do(OopClosure* f) { // tty->print_cr("START OopTaskQueue::oops_do"); - int iters = size(); - juint index = _bottom; - for (int i = 0; i < iters; ++i) { + uint iters = size(); + uint index = _bottom; + for (uint i = 0; i < iters; ++i) { index = decrement_index(index); // tty->print_cr(" doing entry %d," INTPTR_T " -> " INTPTR_T, // index, &_elems[index], _elems[index]); @@ -198,10 +207,10 @@ template<class E> -bool GenericTaskQueue<E>::push_slow(E t, juint dirty_n_elems) { +bool GenericTaskQueue<E>::push_slow(E t, uint dirty_n_elems) { if (dirty_n_elems == n() - 1) { // Actually means 0, so do the push. - juint localBot = _bottom; + uint localBot = _bottom; _elems[localBot] = t; _bottom = increment_index(localBot); return true; @@ -211,7 +220,7 @@ template<class E> bool GenericTaskQueue<E>:: -pop_local_slow(juint localBot, Age oldAge) { +pop_local_slow(uint localBot, Age oldAge) { // This queue was observed to contain exactly one element; either this // thread will claim it, or a competing "pop_global". In either case, // the queue will be logically empty afterwards. Create a new Age value @@ -230,9 +239,8 @@ Age tempAge; // No competing pop_global has yet incremented "top"; we'll try to // install new_age, thus claiming the element. - assert(sizeof(Age) == sizeof(jint) && sizeof(jint) == sizeof(juint), - "Assumption about CAS unit."); - *(jint*)&tempAge = Atomic::cmpxchg(*(jint*)&newAge, (volatile jint*)&_age, *(jint*)&oldAge); + assert(sizeof(Age) == sizeof(int), "Assumption about CAS unit."); + *(uint*)&tempAge = Atomic::cmpxchg(*(uint*)&newAge, (volatile uint*)&_age, *(uint*)&oldAge); if (tempAge == oldAge) { // We win. assert(dirty_size(localBot, get_top()) != n() - 1, @@ -253,8 +261,8 @@ bool GenericTaskQueue<E>::pop_global(E& t) { Age newAge; Age oldAge = get_age(); - juint localBot = _bottom; - juint n_elems = size(localBot, oldAge.top()); + uint localBot = _bottom; + uint n_elems = size(localBot, oldAge.top()); if (n_elems == 0) { return false; } @@ -263,7 +271,7 @@ newAge._top = increment_index(newAge.top()); if ( newAge._top == 0 ) newAge._tag++; Age resAge; - *(jint*)&resAge = Atomic::cmpxchg(*(jint*)&newAge, (volatile jint*)&_age, *(jint*)&oldAge); + *(uint*)&resAge = Atomic::cmpxchg(*(uint*)&newAge, (volatile uint*)&_age, *(uint*)&oldAge); // Note that using "_bottom" here might fail, since a pop_local might // have decremented it. assert(dirty_size(localBot, newAge._top) != n() - 1, @@ -287,7 +295,7 @@ template<class E> class GenericTaskQueueSet: public TaskQueueSetSuper { private: - int _n; + uint _n; GenericTaskQueue<E>** _queues; public: @@ -300,51 +308,51 @@ } } - bool steal_1_random(int queue_num, int* seed, E& t); - bool steal_best_of_2(int queue_num, int* seed, E& t); - bool steal_best_of_all(int queue_num, int* seed, E& t); + bool steal_1_random(uint queue_num, int* seed, E& t); + bool steal_best_of_2(uint queue_num, int* seed, E& t); + bool steal_best_of_all(uint queue_num, int* seed, E& t); - void register_queue(int i, GenericTaskQueue<E>* q); + void register_queue(uint i, GenericTaskQueue<E>* q); - GenericTaskQueue<E>* queue(int n); + GenericTaskQueue<E>* queue(uint n); // The thread with queue number "queue_num" (and whose random number seed // is at "seed") is trying to steal a task from some other queue. (It // may try several queues, according to some configuration parameter.) // If some steal succeeds, returns "true" and sets "t" the stolen task, // otherwise returns false. - bool steal(int queue_num, int* seed, E& t); + bool steal(uint queue_num, int* seed, E& t); bool peek(); }; template<class E> -void GenericTaskQueueSet<E>::register_queue(int i, GenericTaskQueue<E>* q) { - assert(0 <= i && i < _n, "index out of range."); +void GenericTaskQueueSet<E>::register_queue(uint i, GenericTaskQueue<E>* q) { + assert(i < _n, "index out of range."); _queues[i] = q; } template<class E> -GenericTaskQueue<E>* GenericTaskQueueSet<E>::queue(int i) { +GenericTaskQueue<E>* GenericTaskQueueSet<E>::queue(uint i) { return _queues[i]; } template<class E> -bool GenericTaskQueueSet<E>::steal(int queue_num, int* seed, E& t) { - for (int i = 0; i < 2 * _n; i++) +bool GenericTaskQueueSet<E>::steal(uint queue_num, int* seed, E& t) { + for (uint i = 0; i < 2 * _n; i++) if (steal_best_of_2(queue_num, seed, t)) return true; return false; } template<class E> -bool GenericTaskQueueSet<E>::steal_best_of_all(int queue_num, int* seed, E& t) { +bool GenericTaskQueueSet<E>::steal_best_of_all(uint queue_num, int* seed, E& t) { if (_n > 2) { int best_k; - jint best_sz = 0; - for (int k = 0; k < _n; k++) { + uint best_sz = 0; + for (uint k = 0; k < _n; k++) { if (k == queue_num) continue; - jint sz = _queues[k]->size(); + uint sz = _queues[k]->size(); if (sz > best_sz) { best_sz = sz; best_k = k; @@ -362,9 +370,9 @@ } template<class E> -bool GenericTaskQueueSet<E>::steal_1_random(int queue_num, int* seed, E& t) { +bool GenericTaskQueueSet<E>::steal_1_random(uint queue_num, int* seed, E& t) { if (_n > 2) { - int k = queue_num; + uint k = queue_num; while (k == queue_num) k = randomParkAndMiller(seed) % _n; return _queues[2]->pop_global(t); } else if (_n == 2) { @@ -378,20 +386,20 @@ } template<class E> -bool GenericTaskQueueSet<E>::steal_best_of_2(int queue_num, int* seed, E& t) { +bool GenericTaskQueueSet<E>::steal_best_of_2(uint queue_num, int* seed, E& t) { if (_n > 2) { - int k1 = queue_num; + uint k1 = queue_num; while (k1 == queue_num) k1 = randomParkAndMiller(seed) % _n; - int k2 = queue_num; + uint k2 = queue_num; while (k2 == queue_num || k2 == k1) k2 = randomParkAndMiller(seed) % _n; // Sample both and try the larger. - juint sz1 = _queues[k1]->size(); - juint sz2 = _queues[k2]->size(); + uint sz1 = _queues[k1]->size(); + uint sz2 = _queues[k2]->size(); if (sz2 > sz1) return _queues[k2]->pop_global(t); else return _queues[k1]->pop_global(t); } else if (_n == 2) { // Just try the other one. - int k = (queue_num + 1) % 2; + uint k = (queue_num + 1) % 2; return _queues[k]->pop_global(t); } else { assert(_n == 1, "can't be zero."); @@ -402,7 +410,7 @@ template<class E> bool GenericTaskQueueSet<E>::peek() { // Try all the queues. - for (int j = 0; j < _n; j++) { + for (uint j = 0; j < _n; j++) { if (_queues[j]->peek()) return true; } @@ -418,11 +426,19 @@ // A class to aid in the termination of a set of parallel tasks using // TaskQueueSet's for work stealing. +#undef TRACESPINNING + class ParallelTaskTerminator: public StackObj { private: int _n_threads; TaskQueueSetSuper* _queue_set; - jint _offered_termination; + int _offered_termination; + +#ifdef TRACESPINNING + static uint _total_yields; + static uint _total_spins; + static uint _total_peeks; +#endif bool peek_in_queue_set(); protected: @@ -454,13 +470,19 @@ // the terminator is finished. void reset_for_reuse(); +#ifdef TRACESPINNING + static uint total_yields() { return _total_yields; } + static uint total_spins() { return _total_spins; } + static uint total_peeks() { return _total_peeks; } + static void print_termination_counts(); +#endif }; #define SIMPLE_STACK 0 template<class E> inline bool GenericTaskQueue<E>::push(E t) { #if SIMPLE_STACK - juint localBot = _bottom; + uint localBot = _bottom; if (_bottom < max_elems()) { _elems[localBot] = t; _bottom = localBot + 1; @@ -469,10 +491,10 @@ return false; } #else - juint localBot = _bottom; + uint localBot = _bottom; assert((localBot >= 0) && (localBot < n()), "_bottom out of range."); - jushort top = get_top(); - juint dirty_n_elems = dirty_size(localBot, top); + TAG_TYPE top = get_top(); + uint dirty_n_elems = dirty_size(localBot, top); assert((dirty_n_elems >= 0) && (dirty_n_elems < n()), "n_elems out of range."); if (dirty_n_elems < max_elems()) { @@ -487,19 +509,19 @@ template<class E> inline bool GenericTaskQueue<E>::pop_local(E& t) { #if SIMPLE_STACK - juint localBot = _bottom; + uint localBot = _bottom; assert(localBot > 0, "precondition."); localBot--; t = _elems[localBot]; _bottom = localBot; return true; #else - juint localBot = _bottom; + uint localBot = _bottom; // This value cannot be n-1. That can only occur as a result of // the assignment to bottom in this method. If it does, this method // resets the size( to 0 before the next call (which is sequential, // since this is pop_local.) - juint dirty_n_elems = dirty_size(localBot, get_top()); + uint dirty_n_elems = dirty_size(localBot, get_top()); assert(dirty_n_elems != n() - 1, "Shouldn't be possible..."); if (dirty_n_elems == 0) return false; localBot = decrement_index(localBot); @@ -512,7 +534,7 @@ // If there's still at least one element in the queue, based on the // "_bottom" and "age" we've read, then there can be no interference with // a "pop_global" operation, and we're done. - juint tp = get_top(); + TAG_TYPE tp = get_top(); // XXX if (size(localBot, tp) > 0) { assert(dirty_size(localBot, tp) != n() - 1, "Shouldn't be possible..."); @@ -581,7 +603,7 @@ bool is_empty(); bool stealable_is_empty(); bool overflow_is_empty(); - juint stealable_size() { return _region_queue.size(); } + uint stealable_size() { return _region_queue.size(); } RegionTaskQueue* task_queue() { return &_region_queue; } };
--- a/src/share/vm/utilities/workgroup.hpp Thu Feb 19 14:08:00 2009 -0800 +++ b/src/share/vm/utilities/workgroup.hpp Wed Feb 25 23:16:22 2009 -0800 @@ -32,7 +32,7 @@ // An abstract task to be worked on by a gang. // You subclass this to supply your own work() method -class AbstractGangTask: public CHeapObj { +class AbstractGangTask VALUE_OBJ_CLASS_SPEC { public: // The abstract work method. // The argument tells you which member of the gang you are.
--- a/test/Makefile Thu Feb 19 14:08:00 2009 -0800 +++ b/test/Makefile Wed Feb 25 23:16:22 2009 -0800 @@ -28,9 +28,9 @@ # Get OS/ARCH specifics OSNAME = $(shell uname -s) -SLASH_JAVA = /java ifeq ($(OSNAME), SunOS) PLATFORM = solaris + SLASH_JAVA = /java ARCH = $(shell uname -p) ifeq ($(ARCH), i386) ARCH=i586 @@ -38,6 +38,7 @@ endif ifeq ($(OSNAME), Linux) PLATFORM = linux + SLASH_JAVA = /java ARCH = $(shell uname -m) ifeq ($(ARCH), i386) ARCH = i586 @@ -62,6 +63,10 @@ EXESUFFIX = .exe endif +ifdef ALT_SLASH_JAVA + SLASH_JAVA = $(ALT_SLASH_JAVA) +endif + # Utilities used CD = cd CP = cp
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6603011/Test.java Wed Feb 25 23:16:22 2009 -0800 @@ -0,0 +1,220 @@ +/* + * 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 6603011 + * @summary long/int division by constant + * + * @run main/othervm -Xcomp -Xbatch -XX:-Inline Test + */ + +// +// -XX:-Inline is essential to this test so that verification functions +// divi, modi, divl and modl generate "plain" divides. +// -Xcomp -Xbatch are also useful to ensure the full range of +// dividend and divisor combinations are tested +// + +import java.net.*; + +class s { + static int divi(int dividend, int divisor) { return dividend / divisor; } + static int modi(int dividend, int divisor) { return dividend % divisor; } + static long divl(long dividend, long divisor) { return dividend / divisor; } + static long modl(long dividend, long divisor) { return dividend % divisor; } +} + +public class Test implements Runnable { + // Report verbose messages on failure; turn off to suppress + // too much output with gross numbers of failures. + static final boolean VERBOSE = true; + + // Initailize DIVISOR so that it is final in this class. + static final int DIVISOR; + static { + int value = 0; + try { + value = Integer.decode(System.getProperty("divisor")); + } catch (Throwable e) { + } + DIVISOR = value; + } + + // The methods of interest. We want the JIT to compile these + // and convert the divide into a multiply. + public int divbyI (int dividend) { return dividend / DIVISOR; } + public int modbyI (int dividend) { return dividend % DIVISOR; } + public long divbyL (long dividend) { return dividend / DIVISOR; } + public long modbyL (long dividend) { return dividend % DIVISOR; } + + public int divisor() { return DIVISOR; } + + public boolean checkI (int dividend) { + int quo = divbyI(dividend); + int rem = modbyI(dividend); + int quo0 = s.divi(dividend, divisor()); + int rem0 = s.modi(dividend, divisor()); + + if (quo != quo0 || rem != rem0) { + if (VERBOSE) { + System.out.println("Computed: " + dividend + " / " + divisor() + " = " + + quo + ", " + dividend + " % " + divisor() + " = " + rem ); + System.out.println("expected: " + dividend + " / " + divisor() + " = " + + quo0 + ", " + dividend + " % " + divisor() + " = " + rem0); + // Report sign of rem failure + if (rem != 0 && (rem ^ dividend) < 0) { + System.out.println(" rem & dividend have different signs"); + } + // Report range of rem failure + if (java.lang.Math.abs(rem) >= java.lang.Math.abs(divisor())) { + System.out.println(" remainder out of range"); + } + // Report quo/rem identity relationship failure + if ((quo * divisor()) + rem != dividend) { + System.out.println(" quotien/remainder invariant broken"); + } + } + return false; + } + return true; + } + + public boolean checkL (long dividend) { + long quo = divbyL(dividend); + long rem = modbyL(dividend); + long quo0 = s.divl(dividend, divisor()); + long rem0 = s.modl(dividend, divisor()); + + if (quo != quo0 || rem != rem0) { + if (VERBOSE) { + System.out.println(" " + dividend + " / " + divisor() + " = " + + quo + ", " + dividend + " % " + divisor() + " = " + rem); + // Report sign of rem failure + if (rem != 0 && (rem ^ dividend) < 0) { + System.out.println(" rem & dividend have different signs"); + } + // Report range of rem failure + if (java.lang.Math.abs(rem) >= java.lang.Math.abs(divisor())) { + System.out.println(" remainder out of range"); + } + // Report quo/rem identity relationship failure + if ((quo * divisor()) + rem != dividend) { + System.out.println(" (" + quo + " * " + divisor() + ") + " + rem + " != " + + dividend); + } + } + return false; + } + return true; + } + + public void run() { + // Don't try to divide by zero + if (divisor() == 0) return; + + // Range of dividends to check. Try dividends from start to end + // inclusive, as well as variations on those values as shifted + // left. + int start = -1024; + int end = 1024; + + // Test int division using a variety of dividends. + int wrong = 0; + int total = 0; + + outerloop: + for (int i = start; i <= end; i++) { + for (int s = 0; s < 32; s += 4) { + total++; + int dividend = i << s; + if (!checkI(dividend)) { + wrong++; + // Stop on the first failure + // break outerloop; + } + } + } + if (wrong > 0) { + System.out.println("divisor " + divisor() + ": " + + wrong + "/" + total + " wrong int divisions"); + } + + // Test long division using a variety of dividends. + wrong = 0; + total = 0; + + outerloop: + for (int i = start; i <= end; i++) { + for (int s = 0; s < 64; s += 4) { + total++; + long dividend = i << s; + if (!checkL(dividend)) { + wrong++; + // Stop on the first failure + // break outerloop; + } + } + } + if (wrong > 0) { + System.out.println("divisor " + divisor() + ": " + + wrong + "/" + total + " wrong long divisions"); + } + + } + + // Reload this class with the "divisor" property set to the input parameter. + // This allows the JIT to see q.DIVISOR as a final constant, and change + // any divisions or mod operations into multiplies. + public static void test_divisor(int divisor, + URLClassLoader apploader) throws Exception { + System.setProperty("divisor", "" + divisor); + ClassLoader loader = new URLClassLoader(apploader.getURLs(), + apploader.getParent()); + Class c = loader.loadClass("Test"); + Runnable r = (Runnable)c.newInstance(); + r.run(); + } + + public static void main(String[] args) throws Exception { + Class cl = Class.forName("Test"); + URLClassLoader apploader = (URLClassLoader)cl.getClassLoader(); + + + // Test every divisor between -100 and 100. + for (int i = -100; i <= 100; i++) { + test_divisor(i, apploader); + } + + // Try a few divisors outside the typical range. + // The values below have been observed in rt.jar. + test_divisor(101, apploader); + test_divisor(400, apploader); + test_divisor(1000, apploader); + test_divisor(3600, apploader); + test_divisor(9973, apploader); + test_divisor(86400, apploader); + test_divisor(1000000, apploader); + } + +}
--- a/test/compiler/6775880/Test.java Thu Feb 19 14:08:00 2009 -0800 +++ b/test/compiler/6775880/Test.java Wed Feb 25 23:16:22 2009 -0800 @@ -27,7 +27,7 @@ * @bug 6775880 * @summary EA +DeoptimizeALot: assert(mon_info->owner()->is_locked(),"object must be locked now") * @compile -source 1.4 -target 1.4 Test.java - * @run main/othervm -server -Xbatch -XX:+DoEscapeAnalysis -XX:+DeoptimizeALot -XX:CompileCommand=exclude,java.lang.AbstractStringBuilder::append Test + * @run main/othervm -XX:+IgnoreUnrecognizedVMOptions -Xbatch -XX:+DoEscapeAnalysis -XX:+DeoptimizeALot -XX:CompileCommand=exclude,java.lang.AbstractStringBuilder::append Test */ public class Test {
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6795161/Test.java Wed Feb 25 23:16:22 2009 -0800 @@ -0,0 +1,60 @@ +/* + * 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 6795161 + * @summary Escape analysis leads to data corruption + * @run main/othervm -server -Xcomp -XX:CompileOnly=Test -XX:+DoEscapeAnalysis Test + */ + +class Test_Class_1 { + static String var_1; + + static void badFunc(int size) + { + try { + for (int i = 0; i < 1; (new byte[size-i])[0] = 0, i++) {} + } catch (Exception e) { + // don't comment it out, it will lead to correct results ;) + //System.out.println("Got exception: " + e); + } + } +} + +public class Test { + static String var_1_copy = Test_Class_1.var_1; + + static byte var_check; + + public static void main(String[] args) + { + var_check = 1; + + Test_Class_1.badFunc(-1); + + System.out.println("EATester.var_check = " + Test.var_check + " (expected 1)\n"); + } +} +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6795362/Test6795362.java Wed Feb 25 23:16:22 2009 -0800 @@ -0,0 +1,48 @@ +/* + * 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 6795362 + * @summary 32bit server compiler leads to wrong results on solaris-x86 + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6795362.sub Test6795362 + */ + +public class Test6795362 { + public static void main(String[] args) + { + sub(); + + if (var_bad != 0) + throw new InternalError(var_bad + " != 0"); + } + + static long var_bad = -1L; + + static void sub() + { + var_bad >>= 65; + var_bad /= 65; + } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6799693/Test.java Wed Feb 25 23:16:22 2009 -0800 @@ -0,0 +1,47 @@ +/* + * 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 6799693 + * @summary Server compiler leads to data corruption when expression throws an Exception + * @run main/othervm -Xcomp -XX:CompileOnly=Test Test + */ + +public class Test { + static int var_bad = 1; + + public static void main(String[] args) + { + var_bad++; + + try { + for (int i = 0; i < 10; i++) (new byte[((byte)-1 << i)])[0] = 0; + } + catch (Exception e) { System.out.println("Got " + e); } + + System.out.println("Test.var_bad = " + var_bad + " (expected 2)\n"); + } +} +
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6800154/Test6800154.java Wed Feb 25 23:16:22 2009 -0800 @@ -0,0 +1,109 @@ +/* + * 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 6800154 + * @summary Add comments to long_by_long_mulhi() for better understandability + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6800154.divcomp Test6800154 + */ + +import java.net.URLClassLoader; + +public class Test6800154 implements Runnable { + static final long[] DIVIDENDS = { + 0, + 1, + 2, + 1423487, + 4444441, + 4918923241323L, + -1, + -24351, + 0x3333, + 0x0000000080000000L, + 0x7fffffffffffffffL, + 0x8000000000000000L + }; + + static final long[] DIVISORS = { + 1, + 2, + 17, + 12342, + 24123, + 143444, + 123444442344L, + -1, + -2, + -4423423234231423L, + 0x0000000080000000L, + 0x7fffffffffffffffL, + 0x8000000000000000L + }; + + // Initialize DIVISOR so that it is final in this class. + static final long DIVISOR; + + static { + long value = 0; + try { + value = Long.decode(System.getProperty("divisor")); + } catch (Throwable e) { + } + DIVISOR = value; + } + + public static void main(String[] args) throws Exception + { + Class cl = Class.forName("Test6800154"); + URLClassLoader apploader = (URLClassLoader) cl.getClassLoader(); + + // Iterate over all divisors. + for (int i = 0; i < DIVISORS.length; i++) { + System.setProperty("divisor", "" + DIVISORS[i]); + ClassLoader loader = new URLClassLoader(apploader.getURLs(), apploader.getParent()); + Class c = loader.loadClass("Test6800154"); + Runnable r = (Runnable) c.newInstance(); + r.run(); + } + } + + public void run() + { + // Iterate over all dividends. + for (int i = 0; i < DIVIDENDS.length; i++) { + long dividend = DIVIDENDS[i]; + + long expected = divint(dividend); + long result = divcomp(dividend); + + if (result != expected) + throw new InternalError(dividend + " / " + DIVISOR + " failed: " + result + " != " + expected); + } + } + + static long divint(long a) { return a / DIVISOR; } + static long divcomp(long a) { return a / DIVISOR; } +}
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/test/compiler/6805724/Test6805724.java Wed Feb 25 23:16:22 2009 -0800 @@ -0,0 +1,80 @@ +/* + * 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 6805724 + * @summary ModLNode::Ideal() generates functionally incorrect graph when divisor is any (2^k-1) constant. + * + * @run main/othervm -Xcomp -XX:CompileOnly=Test6805724.fcomp Test6805724 + */ + +import java.net.URLClassLoader; + +public class Test6805724 implements Runnable { + // Initialize DIVISOR so that it is final in this class. + static final long DIVISOR; // 2^k-1 constant + + static { + long value = 0; + try { + value = Long.decode(System.getProperty("divisor")); + } catch (Throwable t) { + // This one is required for the Class.forName() in main. + } + DIVISOR = value; + } + + static long fint(long x) { + return x % DIVISOR; + } + + static long fcomp(long x) { + return x % DIVISOR; + } + + public void run() { + long a = 0x617981E1L; + + long expected = fint(a); + long result = fcomp(a); + + if (result != expected) + throw new InternalError(result + " != " + expected); + } + + public static void main(String args[]) throws Exception { + Class cl = Class.forName("Test6805724"); + URLClassLoader apploader = (URLClassLoader) cl.getClassLoader(); + + // Iterate over all 2^k-1 divisors. + for (int k = 1; k < Long.SIZE; k++) { + long divisor = (1L << k) - 1; + System.setProperty("divisor", "" + divisor); + ClassLoader loader = new URLClassLoader(apploader.getURLs(), apploader.getParent()); + Class c = loader.loadClass("Test6805724"); + Runnable r = (Runnable) c.newInstance(); + r.run(); + } + } +}