--- old/src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp 2013-02-20 17:40:21.798721909 +0100 +++ new/src/cpu/sparc/vm/c1_CodeStubs_sparc.cpp 2013-02-20 17:40:21.641345733 +0100 @@ -51,6 +51,15 @@ void RangeCheckStub::emit_code(LIR_Assembler* ce) { __ bind(_entry); + if (_info->deoptimize_on_exception()) { + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(a, relocInfo::runtime_call_type); + __ delayed()->nop(); + ce->add_call_info_here(_info); + debug_only(__ should_not_reach_here()); + return; + } + if (_index->is_register()) { __ mov(_index->as_register(), G4); } else { @@ -69,6 +78,19 @@ #endif } +PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) { + _info = new CodeEmitInfo(info); +} + +void PredicateFailedStub::emit_code(LIR_Assembler* ce) { + __ bind(_entry); + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(a, relocInfo::runtime_call_type); + __ delayed()->nop(); + ce->add_call_info_here(_info); + ce->verify_oop_map(_info); + debug_only(__ should_not_reach_here()); +} void CounterOverflowStub::emit_code(LIR_Assembler* ce) { __ bind(_entry); @@ -99,10 +121,17 @@ void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) { + address a; + if (_info->deoptimize_on_exception()) { + // Deoptimize, do not throw the exception, because it is probably wrong to do it here. + a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + } else { + a = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id); + } + ce->compilation()->implicit_exception_table()->append(_offset, __ offset()); __ bind(_entry); - __ call(Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id), - relocInfo::runtime_call_type); + __ call(a, relocInfo::runtime_call_type); __ delayed()->nop(); ce->add_call_info_here(_info); ce->verify_oop_map(_info); --- old/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp 2013-02-20 17:40:23.052747404 +0100 +++ new/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp 2013-02-20 17:40:22.875985710 +0100 @@ -3362,6 +3362,47 @@ } +#ifndef PRODUCT + +// emit run-time assertion +void LIR_Assembler::emit_assert(LIR_OpAssert* op) { + assert(op->code() == lir_assert, "must be"); + + if (op->in_opr1()->is_valid()) { + assert(op->in_opr2()->is_valid(), "both operands must be valid"); + comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op); + } else { + assert(op->in_opr2()->is_illegal(), "both operands must be illegal"); + assert(op->condition() == lir_cond_always, "no other conditions allowed"); + } + + Label ok; + if (op->condition() != lir_cond_always) { + Assembler::Condition acond; + switch (op->condition()) { + case lir_cond_equal: acond = Assembler::equal; break; + case lir_cond_notEqual: acond = Assembler::notEqual; break; + case lir_cond_less: acond = Assembler::less; break; + case lir_cond_lessEqual: acond = Assembler::lessEqual; break; + case lir_cond_greaterEqual: acond = Assembler::greaterEqual; break; + case lir_cond_greater: acond = Assembler::greater; break; + case lir_cond_aboveEqual: acond = Assembler::greaterEqualUnsigned; break; + case lir_cond_belowEqual: acond = Assembler::lessEqualUnsigned; break; + default: ShouldNotReachHere(); + }; + __ br(acond, false, Assembler::pt, ok); + __ delayed()->nop(); + } + if (op->halt()) { + const char* str = __ code_string(op->msg()); + delete op->msg(); + __ stop(str); + } else { + breakpoint(); + } + __ bind(ok); +} +#endif void LIR_Assembler::peephole(LIR_List* lir) { LIR_OpList* inst = lir->instructions_list(); for (int i = 0; i < inst->length(); i++) { --- old/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp 2013-02-20 17:40:24.444022375 +0100 +++ new/src/cpu/sparc/vm/c1_LIRGenerator_sparc.cpp 2013-02-20 17:40:24.288618142 +0100 @@ -324,7 +324,7 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) { assert(x->is_pinned(),""); - bool needs_range_check = true; + bool needs_range_check = x->compute_needs_range_check(); bool use_length = x->length() != NULL; bool obj_store = x->elt_type() == T_ARRAY || x->elt_type() == T_OBJECT; bool needs_store_check = obj_store && (x->value()->as_Constant() == NULL || @@ -339,12 +339,9 @@ array.load_item(); index.load_nonconstant(); - if (use_length) { - needs_range_check = x->compute_needs_range_check(); - if (needs_range_check) { - length.set_instruction(x->length()); - length.load_item(); - } + if (use_length && needs_range_check) { + length.set_instruction(x->length()); + length.load_item(); } if (needs_store_check) { value.load_item(); --- old/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp 2013-02-20 17:40:25.767043768 +0100 +++ new/src/cpu/sparc/vm/c1_Runtime1_sparc.cpp 2013-02-20 17:40:25.598723144 +0100 @@ -987,6 +987,25 @@ break; #endif // INCLUDE_ALL_GCS + case predicate_failed_trap_id: + { + __ set_info("throw_incompatible_class_cast_exception", dont_gc_arguments); + OopMap* oop_map = save_live_registers(sasm); + + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap)); + + oop_maps = new OopMapSet(); + oop_maps->add_gc_map(call_offset, oop_map); + + DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); + assert(deopt_blob != NULL, "deoptimization blob must have been created"); + restore_live_registers(sasm); + __ restore(); + __ br(Assembler::always, false, Assembler::pt, deopt_blob->unpack_with_reexecution(), relocInfo::runtime_call_type); + __ delayed()->nop(); + } + break; + default: { __ set_info("unimplemented entry", dont_gc_arguments); __ save_frame(0); --- old/src/cpu/x86/vm/c1_CodeStubs_x86.cpp 2013-02-20 17:40:27.103156525 +0100 +++ new/src/cpu/x86/vm/c1_CodeStubs_x86.cpp 2013-02-20 17:40:26.938948147 +0100 @@ -101,6 +101,14 @@ void RangeCheckStub::emit_code(LIR_Assembler* ce) { __ bind(_entry); + if (_info->deoptimize_on_exception()) { + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(RuntimeAddress(a)); + ce->add_call_info_here(_info); + debug_only(__ should_not_reach_here()); + return; + } + // pass the array index on stack because all registers must be preserved if (_index->is_cpu_register()) { ce->store_parameter(_index->as_register(), 0); @@ -118,6 +126,18 @@ debug_only(__ should_not_reach_here()); } +PredicateFailedStub::PredicateFailedStub(CodeEmitInfo* info) { + _info = new CodeEmitInfo(info); +} + +void PredicateFailedStub::emit_code(LIR_Assembler* ce) { + __ bind(_entry); + address a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + __ call(RuntimeAddress(a)); + ce->add_call_info_here(_info); + ce->verify_oop_map(_info); + debug_only(__ should_not_reach_here()); +} void DivByZeroStub::emit_code(LIR_Assembler* ce) { if (_offset != -1) { @@ -414,9 +434,18 @@ void ImplicitNullCheckStub::emit_code(LIR_Assembler* ce) { + + address a; + if (_info->deoptimize_on_exception()) { + // Deoptimize, do not throw the exception, because it is probably wrong to do it here. + a = Runtime1::entry_for(Runtime1::predicate_failed_trap_id); + } else { + a = Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id); + } + ce->compilation()->implicit_exception_table()->append(_offset, __ offset()); __ bind(_entry); - __ call(RuntimeAddress(Runtime1::entry_for(Runtime1::throw_null_pointer_exception_id))); + __ call(RuntimeAddress(a)); ce->add_call_info_here(_info); debug_only(__ should_not_reach_here()); } --- old/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp 2013-02-20 17:40:28.374382330 +0100 +++ new/src/cpu/x86/vm/c1_LIRAssembler_x86.cpp 2013-02-20 17:40:28.186795830 +0100 @@ -3756,6 +3756,47 @@ } +#ifndef PRODUCT + +// emit run-time assertion +void LIR_Assembler::emit_assert(LIR_OpAssert* op) { + assert(op->code() == lir_assert, "must be"); + + if (op->in_opr1()->is_valid()) { + assert(op->in_opr2()->is_valid(), "both operands must be valid"); + comp_op(op->condition(), op->in_opr1(), op->in_opr2(), op); + } else { + assert(op->in_opr2()->is_illegal(), "both operands must be illegal"); + assert(op->condition() == lir_cond_always, "no other conditions allowed"); + } + + Label ok; + if (op->condition() != lir_cond_always) { + Assembler::Condition acond = Assembler::zero; + switch (op->condition()) { + case lir_cond_equal: acond = Assembler::equal; break; + case lir_cond_notEqual: acond = Assembler::notEqual; break; + case lir_cond_less: acond = Assembler::less; break; + case lir_cond_lessEqual: acond = Assembler::lessEqual; break; + case lir_cond_greaterEqual: acond = Assembler::greaterEqual;break; + case lir_cond_greater: acond = Assembler::greater; break; + case lir_cond_belowEqual: acond = Assembler::belowEqual; break; + case lir_cond_aboveEqual: acond = Assembler::aboveEqual; break; + default: ShouldNotReachHere(); + } + __ jcc(acond, ok); + } + if (op->halt()) { + const char* str = __ code_string(op->msg()); + delete op->msg(); + __ stop(str); + } else { + breakpoint(); + } + __ bind(ok); +} +#endif + void LIR_Assembler::membar() { // QQQ sparc TSO uses this, __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad)); --- old/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp 2013-02-20 17:40:29.829234999 +0100 +++ new/src/cpu/x86/vm/c1_LIRGenerator_x86.cpp 2013-02-20 17:40:29.675440505 +0100 @@ -263,7 +263,7 @@ void LIRGenerator::do_StoreIndexed(StoreIndexed* x) { assert(x->is_pinned(),""); - bool needs_range_check = true; + bool needs_range_check = x->compute_needs_range_check(); bool use_length = x->length() != NULL; bool obj_store = x->elt_type() == T_ARRAY || x->elt_type() == T_OBJECT; bool needs_store_check = obj_store && (x->value()->as_Constant() == NULL || @@ -278,12 +278,10 @@ array.load_item(); index.load_nonconstant(); - if (use_length) { - needs_range_check = x->compute_needs_range_check(); - if (needs_range_check) { - length.set_instruction(x->length()); - length.load_item(); - } + if (use_length && needs_range_check) { + length.set_instruction(x->length()); + length.load_item(); + } if (needs_store_check) { value.load_item(); --- old/src/cpu/x86/vm/c1_LinearScan_x86.cpp 2013-02-20 17:40:31.127358523 +0100 +++ new/src/cpu/x86/vm/c1_LinearScan_x86.cpp 2013-02-20 17:40:30.959554825 +0100 @@ -675,7 +675,8 @@ switch (op2->code()) { case lir_cmp: case lir_cmp_fd2i: - case lir_ucmp_fd2i: { + case lir_ucmp_fd2i: + case lir_assert: { assert(left->is_fpu_register(), "invalid LIR"); assert(right->is_fpu_register(), "invalid LIR"); --- old/src/cpu/x86/vm/c1_Runtime1_x86.cpp 2013-02-20 17:40:32.473706489 +0100 +++ new/src/cpu/x86/vm/c1_Runtime1_x86.cpp 2013-02-20 17:40:32.291201679 +0100 @@ -1807,6 +1807,24 @@ break; #endif // INCLUDE_ALL_GCS + case predicate_failed_trap_id: + { + StubFrame f(sasm, "predicate_failed_trap", dont_gc_arguments); + + OopMap* map = save_live_registers(sasm, 1); + + int call_offset = __ call_RT(noreg, noreg, CAST_FROM_FN_PTR(address, predicate_failed_trap)); + oop_maps = new OopMapSet(); + oop_maps->add_gc_map(call_offset, map); + restore_live_registers(sasm); + __ leave(); + DeoptimizationBlob* deopt_blob = SharedRuntime::deopt_blob(); + assert(deopt_blob != NULL, "deoptimization blob must have been created"); + + __ jump(RuntimeAddress(deopt_blob->unpack_with_reexecution())); + } + break; + default: { StubFrame f(sasm, "unimplemented entry", dont_gc_arguments); __ movptr(rax, (int)id); --- old/src/share/vm/c1/c1_Canonicalizer.cpp 2013-02-20 17:40:33.880147732 +0100 +++ new/src/share/vm/c1/c1_Canonicalizer.cpp 2013-02-20 17:40:33.715617210 +0100 @@ -937,4 +937,6 @@ void Canonicalizer::do_ProfileCall(ProfileCall* x) {} void Canonicalizer::do_ProfileInvoke(ProfileInvoke* x) {} void Canonicalizer::do_RuntimeCall(RuntimeCall* x) {} +void Canonicalizer::do_RangeCheckPredicate(RangeCheckPredicate* x) {} +void Canonicalizer::do_Assert(Assert* x) {} void Canonicalizer::do_MemBar(MemBar* x) {} --- old/src/share/vm/c1/c1_Canonicalizer.hpp 2013-02-20 17:40:35.404144141 +0100 +++ new/src/share/vm/c1/c1_Canonicalizer.hpp 2013-02-20 17:40:35.235888329 +0100 @@ -107,6 +107,8 @@ virtual void do_ProfileInvoke (ProfileInvoke* x); virtual void do_RuntimeCall (RuntimeCall* x); virtual void do_MemBar (MemBar* x); + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x); + virtual void do_Assert (Assert* x); }; #endif // SHARE_VM_C1_C1_CANONICALIZER_HPP --- old/src/share/vm/c1/c1_CodeStubs.hpp 2013-02-20 17:40:36.597522856 +0100 +++ new/src/share/vm/c1/c1_CodeStubs.hpp 2013-02-20 17:40:36.424723059 +0100 @@ -166,6 +166,22 @@ #endif // PRODUCT }; +// stub used when predicate fails and deoptimization is needed +class PredicateFailedStub: public CodeStub { + private: + CodeEmitInfo* _info; + + public: + PredicateFailedStub(CodeEmitInfo* info); + virtual void emit_code(LIR_Assembler* e); + virtual CodeEmitInfo* info() const { return _info; } + virtual void visit(LIR_OpVisitState* visitor) { + visitor->do_slow_case(_info); + } +#ifndef PRODUCT + virtual void print_name(outputStream* out) const { out->print("PredicateFailedStub"); } +#endif // PRODUCT +}; class DivByZeroStub: public CodeStub { private: --- old/src/share/vm/c1/c1_Compilation.cpp 2013-02-20 17:40:37.820239479 +0100 +++ new/src/share/vm/c1/c1_Compilation.cpp 2013-02-20 17:40:37.658042984 +0100 @@ -33,13 +33,16 @@ #include "c1/c1_ValueStack.hpp" #include "code/debugInfoRec.hpp" #include "compiler/compileLog.hpp" +#include "c1/c1_RangeCheckElimination.hpp" typedef enum { _t_compile, _t_setup, - _t_optimizeIR, _t_buildIR, + _t_optimize_blocks, + _t_optimize_null_checks, + _t_rangeCheckElimination, _t_emit_lir, _t_linearScan, _t_lirGeneration, @@ -52,7 +55,9 @@ static const char * timer_name[] = { "compile", "setup", - "optimizeIR", + "optimize_blocks", + "optimize_null_checks", + "rangeCheckElimination", "buildIR", "emit_lir", "linearScan", @@ -159,9 +164,9 @@ if (UseC1Optimizations) { NEEDS_CLEANUP // optimization - PhaseTraceTime timeit(_t_optimizeIR); + PhaseTraceTime timeit(_t_optimize_blocks); - _hir->optimize(); + _hir->optimize_blocks(); } _hir->verify(); @@ -180,13 +185,49 @@ _hir->compute_code(); if (UseGlobalValueNumbering) { - ResourceMark rm; + // No resource mark here! LoopInvariantCodeMotion can allocate ValueStack objects. int instructions = Instruction::number_of_instructions(); GlobalValueNumbering gvn(_hir); assert(instructions == Instruction::number_of_instructions(), "shouldn't have created an instructions"); } + _hir->verify(); + +#ifndef PRODUCT + + if (PrintCFGToFile) { + CFGPrinter::print_cfg(_hir, "Before RangeCheckElimination", true, false); + } + +#endif + + if (RangeCheckElimination) { + if (_hir->osr_entry() == NULL) { + PhaseTraceTime timeit(_t_rangeCheckElimination); + RangeCheckElimination::eliminate(_hir); + } + } + +#ifndef PRODUCT + if (PrintCFGToFile) { + CFGPrinter::print_cfg(_hir, "After RangeCheckElimination", true, false); + } +#endif + + if (UseC1Optimizations) { + // loop invariant code motion reorders instructions and range + // check elimination adds new instructions so do null check + // elimination after. + NEEDS_CLEANUP + // optimization + PhaseTraceTime timeit(_t_optimize_null_checks); + + _hir->eliminate_null_checks(); + } + + _hir->verify(); + // compute use counts after global value numbering _hir->compute_use_counts(); @@ -502,6 +543,7 @@ , _next_id(0) , _next_block_id(0) , _code(buffer_blob) +, _has_access_indexed(false) , _current_instruction(NULL) #ifndef PRODUCT , _last_instruction_printed(NULL) @@ -567,7 +609,9 @@ tty->print_cr(" Detailed C1 Timings"); tty->print_cr(" Setup time: %6.3f s (%4.1f%%)", timers[_t_setup].seconds(), (timers[_t_setup].seconds() / total) * 100.0); tty->print_cr(" Build IR: %6.3f s (%4.1f%%)", timers[_t_buildIR].seconds(), (timers[_t_buildIR].seconds() / total) * 100.0); - tty->print_cr(" Optimize: %6.3f s (%4.1f%%)", timers[_t_optimizeIR].seconds(), (timers[_t_optimizeIR].seconds() / total) * 100.0); + float t_optimizeIR = timers[_t_optimize_blocks].seconds() + timers[_t_optimize_null_checks].seconds(); + tty->print_cr(" Optimize: %6.3f s (%4.1f%%)", t_optimizeIR, (t_optimizeIR / total) * 100.0); + tty->print_cr(" RCE: %6.3f s (%4.1f%%)", timers[_t_rangeCheckElimination].seconds(), (timers[_t_rangeCheckElimination].seconds() / total) * 100.0); tty->print_cr(" Emit LIR: %6.3f s (%4.1f%%)", timers[_t_emit_lir].seconds(), (timers[_t_emit_lir].seconds() / total) * 100.0); tty->print_cr(" LIR Gen: %6.3f s (%4.1f%%)", timers[_t_lirGeneration].seconds(), (timers[_t_lirGeneration].seconds() / total) * 100.0); tty->print_cr(" Linear Scan: %6.3f s (%4.1f%%)", timers[_t_linearScan].seconds(), (timers[_t_linearScan].seconds() / total) * 100.0); --- old/src/share/vm/c1/c1_Compilation.hpp 2013-02-20 17:40:39.065248470 +0100 +++ new/src/share/vm/c1/c1_Compilation.hpp 2013-02-20 17:40:38.915704517 +0100 @@ -26,8 +26,10 @@ #define SHARE_VM_C1_C1_COMPILATION_HPP #include "ci/ciEnv.hpp" +#include "ci/ciMethodData.hpp" #include "code/exceptionHandlerTable.hpp" #include "memory/resourceArea.hpp" +#include "runtime/deoptimization.hpp" class CompilationResourceObj; class XHandlers; @@ -85,6 +87,7 @@ LinearScan* _allocator; CodeOffsets _offsets; CodeBuffer _code; + bool _has_access_indexed; // compilation helpers void initialize(); @@ -140,6 +143,7 @@ C1_MacroAssembler* masm() const { return _masm; } CodeOffsets* offsets() { return &_offsets; } Arena* arena() { return _arena; } + bool has_access_indexed() { return _has_access_indexed; } // Instruction ids int get_next_id() { return _next_id++; } @@ -154,6 +158,7 @@ void set_has_fpu_code(bool f) { _has_fpu_code = f; } void set_has_unsafe_access(bool f) { _has_unsafe_access = f; } void set_would_profile(bool f) { _would_profile = f; } + void set_has_access_indexed(bool f) { _has_access_indexed = f; } // Add a set of exception handlers covering the given PC offset void add_exception_handlers_for_pco(int pco, XHandlers* exception_handlers); // Statistics gathering @@ -233,6 +238,14 @@ return env()->comp_level() == CompLevel_full_profile && C1UpdateMethodData && C1ProfileCheckcasts; } + + // will compilation make optimistic assumptions that might lead to + // deoptimization and that the runtime will account for? + bool is_optimistic() const { + return !TieredCompilation && + (RangeCheckElimination || UseLoopInvariantCodeMotion) && + method()->method_data()->trap_count(Deoptimization::Reason_none) == 0; + } }; --- old/src/share/vm/c1/c1_GraphBuilder.cpp 2013-02-20 17:40:40.351906603 +0100 +++ new/src/share/vm/c1/c1_GraphBuilder.cpp 2013-02-20 17:40:40.165379379 +0100 @@ -947,7 +947,9 @@ void GraphBuilder::load_indexed(BasicType type) { - ValueStack* state_before = copy_state_for_exception(); + // In case of in block code motion in range check elimination + ValueStack* state_before = compilation()->is_optimistic() ? copy_state_before() : copy_state_for_exception(); + compilation()->set_has_access_indexed(true); Value index = ipop(); Value array = apop(); Value length = NULL; @@ -961,7 +963,9 @@ void GraphBuilder::store_indexed(BasicType type) { - ValueStack* state_before = copy_state_for_exception(); + // In case of in block code motion in range check elimination + ValueStack* state_before = compilation()->is_optimistic() ? copy_state_before() : copy_state_for_exception(); + compilation()->set_has_access_indexed(true); Value value = pop(as_ValueType(type)); Value index = ipop(); Value array = apop(); @@ -1179,7 +1183,9 @@ BlockBegin* tsux = block_at(stream()->get_dest()); BlockBegin* fsux = block_at(stream()->next_bci()); bool is_bb = tsux->bci() < stream()->cur_bci() || fsux->bci() < stream()->cur_bci(); - Instruction *i = append(new If(x, cond, false, y, tsux, fsux, is_bb ? state_before : NULL, is_bb)); + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + Instruction *i = append(new If(x, cond, false, y, tsux, fsux, (is_bb || compilation()->is_optimistic()) ? state_before : NULL, is_bb)); assert(i->as_Goto() == NULL || (i->as_Goto()->sux_at(0) == tsux && i->as_Goto()->is_safepoint() == tsux->bci() < stream()->cur_bci()) || @@ -1294,7 +1300,9 @@ BlockBegin* tsux = block_at(bci() + sw.dest_offset_at(0)); BlockBegin* fsux = block_at(bci() + sw.default_offset()); bool is_bb = tsux->bci() < bci() || fsux->bci() < bci(); - ValueStack* state_before = is_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = (is_bb || compilation()->is_optimistic()) ? copy_state_before() : NULL; append(new If(ipop(), If::eql, true, key, tsux, fsux, state_before, is_bb)); } else { // collect successors @@ -1308,7 +1316,9 @@ // add default successor if (sw.default_offset() < 0) has_bb = true; sux->at_put(i, block_at(bci() + sw.default_offset())); - ValueStack* state_before = has_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = (has_bb || compilation()->is_optimistic())? copy_state_before() : NULL; Instruction* res = append(new TableSwitch(ipop(), sux, sw.low_key(), state_before, has_bb)); #ifdef ASSERT if (res->as_Goto()) { @@ -1336,7 +1346,9 @@ BlockBegin* tsux = block_at(bci() + pair.offset()); BlockBegin* fsux = block_at(bci() + sw.default_offset()); bool is_bb = tsux->bci() < bci() || fsux->bci() < bci(); - ValueStack* state_before = is_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = (is_bb || compilation()->is_optimistic()) ? copy_state_before() : NULL; append(new If(ipop(), If::eql, true, key, tsux, fsux, state_before, is_bb)); } else { // collect successors & keys @@ -1353,7 +1365,9 @@ // add default successor if (sw.default_offset() < 0) has_bb = true; sux->at_put(i, block_at(bci() + sw.default_offset())); - ValueStack* state_before = has_bb ? copy_state_before() : NULL; + // In case of loop invariant code motion or predicate insertion + // before the body of a loop the state is needed + ValueStack* state_before = (has_bb || compilation()->is_optimistic()) ? copy_state_before() : NULL; Instruction* res = append(new LookupSwitch(ipop(), sux, keys, state_before, has_bb)); #ifdef ASSERT if (res->as_Goto()) { --- old/src/share/vm/c1/c1_IR.cpp 2013-02-20 17:40:41.771811963 +0100 +++ new/src/share/vm/c1/c1_IR.cpp 2013-02-20 17:40:41.607678746 +0100 @@ -182,13 +182,14 @@ // Implementation of CodeEmitInfo // Stack must be NON-null -CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers) +CodeEmitInfo::CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception) : _scope(stack->scope()) , _scope_debug_info(NULL) , _oop_map(NULL) , _stack(stack) , _exception_handlers(exception_handlers) - , _is_method_handle_invoke(false) { + , _is_method_handle_invoke(false) + , _deoptimize_on_exception(deoptimize_on_exception) { assert(_stack != NULL, "must be non null"); } @@ -199,7 +200,8 @@ , _scope_debug_info(NULL) , _oop_map(NULL) , _stack(stack == NULL ? info->_stack : stack) - , _is_method_handle_invoke(info->_is_method_handle_invoke) { + , _is_method_handle_invoke(info->_is_method_handle_invoke) + , _deoptimize_on_exception(info->_deoptimize_on_exception) { // deep copy of exception handlers if (info->_exception_handlers != NULL) { @@ -239,7 +241,7 @@ } -void IR::optimize() { +void IR::optimize_blocks() { Optimizer opt(this); if (!compilation()->profile_branches()) { if (DoCEE) { @@ -257,6 +259,10 @@ #endif } } +} + +void IR::eliminate_null_checks() { + Optimizer opt(this); if (EliminateNullChecks) { opt.eliminate_null_checks(); #ifndef PRODUCT @@ -429,6 +435,7 @@ BlockList _loop_end_blocks; // list of all loop end blocks collected during count_edges BitMap2D _loop_map; // two-dimensional bit set: a bit is set if a block is contained in a loop BlockList _work_list; // temporary list (used in mark_loops and compute_order) + BlockList _loop_headers; Compilation* _compilation; @@ -594,6 +601,7 @@ TRACE_LINEAR_SCAN(3, tty->print_cr("Block B%d is loop header of loop %d", cur->block_id(), _num_loops)); cur->set_loop_index(_num_loops); + _loop_headers.append(cur); _num_loops++; } @@ -656,6 +664,17 @@ // -> this is not a natural loop, so ignore it TRACE_LINEAR_SCAN(2, tty->print_cr("Loop %d is non-natural, so it is ignored", i)); + BlockBegin *loop_header = _loop_headers.at(i); + assert(loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Must be loop header"); + + for (int j = 0; j < loop_header->number_of_preds(); j++) + { + BlockBegin *pred = loop_header->pred_at(j); + pred->clear(BlockBegin::linear_scan_loop_end_flag); + } + + loop_header->clear(BlockBegin::linear_scan_loop_header_flag); + for (int block_id = _max_block_id - 1; block_id >= 0; block_id--) { clear_block_in_loop(i, block_id); } @@ -729,9 +748,20 @@ } else if (!(cur->is_set(BlockBegin::linear_scan_loop_header_flag) && parent->is_set(BlockBegin::linear_scan_loop_end_flag))) { TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: computing dominator of B%d: common dominator of B%d and B%d is B%d", cur->block_id(), parent->block_id(), cur->dominator()->block_id(), common_dominator(cur->dominator(), parent)->block_id())); - assert(cur->number_of_preds() > 1, ""); + // Does not hold for exception blocks + assert(cur->number_of_preds() > 1 || cur->is_set(BlockBegin::exception_entry_flag), ""); cur->set_dominator(common_dominator(cur->dominator(), parent)); } + + // Additional edge to xhandler of all our successors + // range check elimination needs that the state at the end of a + // block be valid in every block it dominates so cur must dominate + // the exception handlers of its successors. + int num_cur_xhandler = cur->number_of_exception_handlers(); + for (int j = 0; j < num_cur_xhandler; j++) { + BlockBegin* xhandler = cur->exception_handler_at(j); + compute_dominator(xhandler, parent); + } } @@ -898,7 +928,6 @@ num_sux = cur->number_of_exception_handlers(); for (i = 0; i < num_sux; i++) { BlockBegin* sux = cur->exception_handler_at(i); - compute_dominator(sux, cur); if (ready_for_processing(sux)) { sort_into_work_list(sux); } @@ -918,8 +947,23 @@ BlockBegin* dominator = block->pred_at(0); int num_preds = block->number_of_preds(); - for (int i = 1; i < num_preds; i++) { - dominator = common_dominator(dominator, block->pred_at(i)); + + TRACE_LINEAR_SCAN(4, tty->print_cr("DOM: Processing B%d", block->block_id())); + + for (int j = 0; j < num_preds; j++) { + + BlockBegin *pred = block->pred_at(j); + TRACE_LINEAR_SCAN(4, tty->print_cr(" DOM: Subrocessing B%d", pred->block_id())); + + if (block->is_set(BlockBegin::exception_entry_flag)) { + dominator = common_dominator(dominator, pred); + int num_pred_preds = pred->number_of_preds(); + for (int k = 0; k < num_pred_preds; k++) { + dominator = common_dominator(dominator, pred->pred_at(k)); + } + } else { + dominator = common_dominator(dominator, pred); + } } if (dominator != block->dominator()) { @@ -946,6 +990,21 @@ // check that dominators are correct assert(!compute_dominators_iter(), "fix point not reached"); + + // Add Blocks to dominates-Array + int num_blocks = _linear_scan_order->length(); + for (int i = 0; i < num_blocks; i++) { + BlockBegin* block = _linear_scan_order->at(i); + + BlockBegin *dom = block->dominator(); + if (dom) { + assert(dom->dominator_depth() != -1, "Dominator must have been visited before"); + dom->dominates()->append(block); + block->set_dominator_depth(dom->dominator_depth() + 1); + } else { + block->set_dominator_depth(0); + } + } } @@ -1032,7 +1091,7 @@ BlockBegin* sux = cur->sux_at(j); assert(sux->linear_scan_number() >= 0 && sux->linear_scan_number() == _linear_scan_order->index_of(sux), "incorrect linear_scan_number"); - if (!cur->is_set(BlockBegin::linear_scan_loop_end_flag)) { + if (!sux->is_set(BlockBegin::backward_branch_target_flag)) { assert(cur->linear_scan_number() < sux->linear_scan_number(), "invalid order"); } if (cur->loop_depth() == sux->loop_depth()) { @@ -1044,7 +1103,7 @@ BlockBegin* pred = cur->pred_at(j); assert(pred->linear_scan_number() >= 0 && pred->linear_scan_number() == _linear_scan_order->index_of(pred), "incorrect linear_scan_number"); - if (!cur->is_set(BlockBegin::linear_scan_loop_header_flag)) { + if (!cur->is_set(BlockBegin::backward_branch_target_flag)) { assert(cur->linear_scan_number() > pred->linear_scan_number(), "invalid order"); } if (cur->loop_depth() == pred->loop_depth()) { @@ -1060,7 +1119,8 @@ } else { assert(cur->dominator() != NULL, "all but first block must have dominator"); } - assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0), "Single predecessor must also be dominator"); + // Assertion does not hold for exception handlers + assert(cur->number_of_preds() != 1 || cur->dominator() == cur->pred_at(0) || cur->is_set(BlockBegin::exception_entry_flag), "Single predecessor must also be dominator"); } // check that all loops are continuous @@ -1249,9 +1309,25 @@ } }; +class VerifyBlockBeginField : public BlockClosure { + +public: + + virtual void block_do(BlockBegin *block) { + + Instruction *cur = block; + while (cur) { + assert(cur->block() == block, "Block begin is not correct"); + cur = cur->next(); + } + } +}; + void IR::verify() { #ifdef ASSERT PredecessorValidator pv(this); + VerifyBlockBeginField verifier; + this->iterate_postorder(&verifier); #endif } --- old/src/share/vm/c1/c1_IR.hpp 2013-02-20 17:40:43.134603734 +0100 +++ new/src/share/vm/c1/c1_IR.hpp 2013-02-20 17:40:42.965341475 +0100 @@ -254,6 +254,7 @@ OopMap* _oop_map; ValueStack* _stack; // used by deoptimization (contains also monitors bool _is_method_handle_invoke; // true if the associated call site is a MethodHandle call site. + bool _deoptimize_on_exception; FrameMap* frame_map() const { return scope()->compilation()->frame_map(); } Compilation* compilation() const { return scope()->compilation(); } @@ -261,7 +262,7 @@ public: // use scope from ValueStack - CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers); + CodeEmitInfo(ValueStack* stack, XHandlers* exception_handlers, bool deoptimize_on_exception = false); // make a copy CodeEmitInfo(CodeEmitInfo* info, ValueStack* stack = NULL); @@ -272,6 +273,7 @@ IRScope* scope() const { return _scope; } XHandlers* exception_handlers() const { return _exception_handlers; } ValueStack* stack() const { return _stack; } + bool deoptimize_on_exception() const { return _deoptimize_on_exception; } void add_register_oop(LIR_Opr opr); void record_debug_info(DebugInformationRecorder* recorder, int pc_offset); @@ -309,7 +311,8 @@ int max_stack() const { return top_scope()->max_stack(); } // expensive // ir manipulation - void optimize(); + void optimize_blocks(); + void eliminate_null_checks(); void compute_predecessors(); void split_critical_edges(); void compute_code(); --- old/src/share/vm/c1/c1_Instruction.cpp 2013-02-20 17:40:44.408461253 +0100 +++ new/src/share/vm/c1/c1_Instruction.cpp 2013-02-20 17:40:44.248644428 +0100 @@ -34,6 +34,15 @@ // Implementation of Instruction +int Instruction::dominator_depth() { + int result = -1; + if (block()) { + result = block()->dominator_depth(); + } + assert(result != -1 || this->as_Local(), "Only locals have dominator depth -1"); + return result; +} + Instruction::Condition Instruction::mirror(Condition cond) { switch (cond) { case eql: return eql; @@ -42,6 +51,8 @@ case leq: return geq; case gtr: return lss; case geq: return leq; + case aeq: return beq; + case beq: return aeq; } ShouldNotReachHere(); return eql; @@ -56,6 +67,8 @@ case leq: return gtr; case gtr: return leq; case geq: return lss; + case aeq: assert(false, "Above equal cannot be negated"); + case beq: assert(false, "Below equal cannot be negated"); } ShouldNotReachHere(); return eql; @@ -70,10 +83,10 @@ } } - -Instruction* Instruction::prev(BlockBegin* block) { +// Prev without need to have BlockBegin +Instruction* Instruction::prev() { Instruction* p = NULL; - Instruction* q = block; + Instruction* q = block(); while (q != this) { assert(q != NULL, "this is not in the block's instruction list"); p = q; q = q->next(); @@ -122,15 +135,24 @@ // perform constant and interval tests on index value bool AccessIndexed::compute_needs_range_check() { - Constant* clength = length()->as_Constant(); - Constant* cindex = index()->as_Constant(); - if (clength && cindex) { - IntConstant* l = clength->type()->as_IntConstant(); - IntConstant* i = cindex->type()->as_IntConstant(); - if (l && i && i->value() < l->value() && i->value() >= 0) { - return false; + + if (length()) { + + Constant* clength = length()->as_Constant(); + Constant* cindex = index()->as_Constant(); + if (clength && cindex) { + IntConstant* l = clength->type()->as_IntConstant(); + IntConstant* i = cindex->type()->as_IntConstant(); + if (l && i && i->value() < l->value() && i->value() >= 0) { + return false; + } } } + + if (!this->check_flag(NeedsRangeCheckFlag)) { + return false; + } + return true; } @@ -631,19 +653,25 @@ // of the inserted block, without recomputing the values of the other blocks // in the CFG. Therefore the value of "depth_first_number" in BlockBegin becomes meaningless. BlockBegin* BlockBegin::insert_block_between(BlockBegin* sux) { - BlockBegin* new_sux = new BlockBegin(end()->state()->bci()); + int bci = sux->bci(); + // critical edge splitting may introduce a goto after a if and array + // bound check elimination may insert a predicate between the if and + // goto. The bci of the goto can't be the one of the if otherwise + // the state and bci are inconsistent and a deoptimization triggered + // by the predicate would lead to incorrect execution/a crash. + BlockBegin* new_sux = new BlockBegin(bci); // mark this block (special treatment when block order is computed) new_sux->set(critical_edge_split_flag); // This goto is not a safepoint. Goto* e = new Goto(sux, false); - new_sux->set_next(e, end()->state()->bci()); + new_sux->set_next(e, bci); new_sux->set_end(e); // setup states ValueStack* s = end()->state(); - new_sux->set_state(s->copy()); - e->set_state(s->copy()); + new_sux->set_state(s->copy(s->kind(), bci)); + e->set_state(s->copy(s->kind(), bci)); assert(new_sux->state()->locals_size() == s->locals_size(), "local size mismatch!"); assert(new_sux->state()->stack_size() == s->stack_size(), "stack size mismatch!"); assert(new_sux->state()->locks_size() == s->locks_size(), "locks size mismatch!"); @@ -960,15 +988,14 @@ BlockList* sux = NULL; if (begin != NULL) { sux = begin->successors(); - } else if (_begin != NULL) { + } else if (this->begin() != NULL) { // copy our sux list - BlockList* sux = new BlockList(_begin->number_of_sux()); - for (int i = 0; i < _begin->number_of_sux(); i++) { - sux->append(_begin->sux_at(i)); + BlockList* sux = new BlockList(this->begin()->number_of_sux()); + for (int i = 0; i < this->begin()->number_of_sux(); i++) { + sux->append(this->begin()->sux_at(i)); } } _sux = sux; - _begin = begin; } @@ -1008,7 +1035,38 @@ } } +#ifndef PRODUCT +// Constructor of Assert +Assert::Assert(Value x, Condition cond, bool unordered_is_true, Value y) : Instruction(illegalType) + , _x(x) + , _cond(cond) + , _y(y) +{ + set_flag(UnorderedIsTrueFlag, unordered_is_true); + assert(x->type()->tag() == y->type()->tag(), "types must match"); + pin(); + + stringStream strStream; + Compilation::current()->method()->print_name(&strStream); + + stringStream strStream1; + InstructionPrinter ip1(1, &strStream1); + ip1.print_instr(x); + + stringStream strStream2; + InstructionPrinter ip2(1, &strStream2); + ip2.print_instr(y); + + _message = new char[1024]; + snprintf(_message, 1024, "Assertion %s %s %s in method %s", strStream1.as_string(), ip2.cond_name(cond), strStream2.as_string(), strStream.as_string()); +} + +#endif + +void RangeCheckPredicate::check_state() { + assert(state()->kind() != ValueStack::EmptyExceptionState && state()->kind() != ValueStack::ExceptionState, "will deopt with empty state"); +} void ProfileInvoke::state_values_do(ValueVisitor* f) { if (state() != NULL) state()->values_do(f); --- old/src/share/vm/c1/c1_Instruction.hpp 2013-02-20 17:40:45.718369068 +0100 +++ new/src/share/vm/c1/c1_Instruction.hpp 2013-02-20 17:40:45.543596265 +0100 @@ -110,6 +110,8 @@ class ProfileInvoke; class RuntimeCall; class MemBar; +class RangeCheckPredicate; +class Assert; // A Value is a reference to the instruction creating the value typedef Instruction* Value; @@ -210,6 +212,10 @@ virtual void do_ProfileInvoke (ProfileInvoke* x) = 0; virtual void do_RuntimeCall (RuntimeCall* x) = 0; virtual void do_MemBar (MemBar* x) = 0; + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x) = 0; +#ifdef ASSERT + virtual void do_Assert (Assert* x) = 0; +#endif }; @@ -306,8 +312,9 @@ void update_exception_state(ValueStack* state); - //protected: - public: + protected: + BlockBegin* _block; // Block that contains this instruction + void set_type(ValueType* type) { assert(type != NULL, "type must exist"); _type = type; @@ -342,6 +349,9 @@ ThrowIncompatibleClassChangeErrorFlag, ProfileMDOFlag, IsLinkedInBlockFlag, + NeedsRangeCheckFlag, + InWorkListFlag, + DeoptimizeOnException, InstructionLastFlag }; @@ -351,7 +361,7 @@ // 'globally' used condition values enum Condition { - eql, neq, lss, leq, gtr, geq + eql, neq, lss, leq, gtr, geq, aeq, beq }; // Instructions may be pinned for many reasons and under certain conditions @@ -381,6 +391,7 @@ , _pin_state(0) , _type(type) , _next(NULL) + , _block(NULL) , _subst(NULL) , _flags(0) , _operand(LIR_OprFact::illegalOpr) @@ -399,11 +410,13 @@ int printable_bci() const { assert(has_printable_bci(), "_printable_bci should have been set"); return _printable_bci; } void set_printable_bci(int bci) { _printable_bci = bci; } #endif + int dominator_depth(); int use_count() const { return _use_count; } int pin_state() const { return _pin_state; } bool is_pinned() const { return _pin_state != 0 || PinAllInstructions; } ValueType* type() const { return _type; } - Instruction* prev(BlockBegin* block); // use carefully, expensive operation + BlockBegin *block() const { return _block; } + Instruction* prev(); // use carefully, expensive operation Instruction* next() const { return _next; } bool has_subst() const { return _subst != NULL; } Instruction* subst() { return _subst == NULL ? this : _subst->subst(); } @@ -432,6 +445,9 @@ assert(as_BlockEnd() == NULL, "BlockEnd instructions must have no next"); assert(next->can_be_linked(), "shouldn't link these instructions into list"); + BlockBegin *block = this->block(); + next->_block = block; + next->set_flag(Instruction::IsLinkedInBlockFlag, true); _next = next; return next; @@ -444,6 +460,29 @@ return set_next(next); } + // when blocks are merged + void fixup_block_pointers() { + Instruction *cur = next()->next(); // next()'s block is set in set_next + while (cur && cur->_block != block()) { + cur->_block = block(); + cur = cur->next(); + } + } + + Instruction *insert_after(Instruction *i) { + Instruction* n = _next; + set_next(i); + i->set_next(n); + return _next; + } + + Instruction *insert_after_same_bci(Instruction *i) { +#ifndef PRODUCT + i->set_printable_bci(printable_bci()); +#endif + return insert_after(i); + } + void set_subst(Instruction* subst) { assert(subst == NULL || type()->base() == subst->type()->base() || @@ -452,6 +491,7 @@ } void set_exception_handlers(XHandlers *xhandlers) { _exception_handlers = xhandlers; } void set_exception_state(ValueStack* s) { check_state(s); _exception_state = s; } + void set_state_before(ValueStack* s) { check_state(s); _state_before = s; } // machine-specifics void set_operand(LIR_Opr operand) { assert(operand != LIR_OprFact::illegalOpr, "operand must exist"); _operand = operand; } @@ -509,6 +549,11 @@ virtual ExceptionObject* as_ExceptionObject() { return NULL; } virtual UnsafeOp* as_UnsafeOp() { return NULL; } virtual ProfileInvoke* as_ProfileInvoke() { return NULL; } + virtual RangeCheckPredicate* as_RangeCheckPredicate() { return NULL; } + +#ifndef PRODUCT + virtual Assert* as_Assert() { return NULL; } +#endif virtual void visit(InstructionVisitor* v) = 0; @@ -570,7 +615,6 @@ LEAF(Phi, Instruction) private: - BlockBegin* _block; // the block to which the phi function belongs int _pf_flags; // the flags of the phi function int _index; // to value on operand stack (index < 0) or to local public: @@ -578,9 +622,9 @@ Phi(ValueType* type, BlockBegin* b, int index) : Instruction(type->base()) , _pf_flags(0) - , _block(b) , _index(index) { + _block = b; NOT_PRODUCT(set_printable_bci(Value(b)->printable_bci())); if (type->is_illegal()) { make_illegal(); @@ -603,8 +647,6 @@ Value operand_at(int i) const; int operand_count() const; - BlockBegin* block() const { return _block; } - void set(Flag f) { _pf_flags |= f; } void clear(Flag f) { _pf_flags &= ~f; } bool is_set(Flag f) const { return (_pf_flags & f) != 0; } @@ -670,6 +712,7 @@ pin(); } + // generic virtual bool can_trap() const { return state_before() != NULL; } virtual void input_values_do(ValueVisitor* f) { /* no values */ } @@ -852,6 +895,7 @@ , _length(length) , _elt_type(elt_type) { + set_flag(Instruction::NeedsRangeCheckFlag, true); ASSERT_VALUES } @@ -860,6 +904,7 @@ Value length() const { return _length; } BasicType elt_type() const { return _elt_type; } + void clear_length() { _length = NULL; } // perform elimination of range checks involving constants bool compute_needs_range_check(); @@ -1524,6 +1569,7 @@ int _bci; // start-bci of block int _depth_first_number; // number of this block in a depth-first ordering int _linear_scan_number; // number of this block in linear-scan ordering + int _dominator_depth; int _loop_depth; // the loop nesting level of this block int _loop_index; // number of the innermost loop of this block int _flags; // the flags associated with this block @@ -1535,6 +1581,7 @@ // SSA specific fields: (factor out later) BlockList _successors; // the successors of this block BlockList _predecessors; // the predecessors of this block + BlockList _dominates; // list of blocks that are dominated by this block BlockBegin* _dominator; // the dominator of this block // SSA specific ends BlockEnd* _end; // the last instruction of this block @@ -1583,10 +1630,12 @@ , _linear_scan_number(-1) , _loop_depth(0) , _flags(0) + , _dominator_depth(-1) , _dominator(NULL) , _end(NULL) , _predecessors(2) , _successors(2) + , _dominates(2) , _exception_handlers(1) , _exception_states(NULL) , _exception_handler_pco(-1) @@ -1603,6 +1652,7 @@ , _total_preds(0) , _stores_to_locals() { + _block = this; #ifndef PRODUCT set_printable_bci(bci); #endif @@ -1612,8 +1662,10 @@ int block_id() const { return _block_id; } int bci() const { return _bci; } BlockList* successors() { return &_successors; } + BlockList* dominates() { return &_dominates; } BlockBegin* dominator() const { return _dominator; } int loop_depth() const { return _loop_depth; } + int dominator_depth() const { return _dominator_depth; } int depth_first_number() const { return _depth_first_number; } int linear_scan_number() const { return _linear_scan_number; } BlockEnd* end() const { return _end; } @@ -1634,6 +1686,7 @@ // manipulation void set_dominator(BlockBegin* dom) { _dominator = dom; } void set_loop_depth(int d) { _loop_depth = d; } + void set_dominator_depth(int d) { _dominator_depth = d; } void set_depth_first_number(int dfn) { _depth_first_number = dfn; } void set_linear_scan_number(int lsn) { _linear_scan_number = lsn; } void set_end(BlockEnd* end); @@ -1695,7 +1748,8 @@ parser_loop_header_flag = 1 << 7, // set by parser to identify blocks where phi functions can not be created on demand critical_edge_split_flag = 1 << 8, // set for all blocks that are introduced when critical edges are split linear_scan_loop_header_flag = 1 << 9, // set during loop-detection for LinearScan - linear_scan_loop_end_flag = 1 << 10 // set during loop-detection for LinearScan + linear_scan_loop_end_flag = 1 << 10, // set during loop-detection for LinearScan + donot_eliminate_range_checks = 1 << 11 // Should be try to eliminate range checks in this block }; void set(Flag f) { _flags |= f; } @@ -1728,7 +1782,6 @@ BASE(BlockEnd, StateSplit) private: - BlockBegin* _begin; BlockList* _sux; protected: @@ -1746,7 +1799,6 @@ // creation BlockEnd(ValueType* type, ValueStack* state_before, bool is_safepoint) : StateSplit(type, state_before) - , _begin(NULL) , _sux(NULL) { set_flag(IsSafepointFlag, is_safepoint); @@ -1754,7 +1806,8 @@ // accessors bool is_safepoint() const { return check_flag(IsSafepointFlag); } - BlockBegin* begin() const { return _begin; } + // For compatibility with old code, for new code use block() + BlockBegin* begin() const { return _block; } // manipulation void set_begin(BlockBegin* begin); @@ -1811,6 +1864,76 @@ void set_direction(Direction d) { _direction = d; } }; +#ifndef PRODUCT + +LEAF(Assert, Instruction) + private: + Value _x; + Condition _cond; + Value _y; + char *_message; + + public: + // creation + // unordered_is_true is valid for float/double compares only + Assert(Value x, Condition cond, bool unordered_is_true, Value y); + + // accessors + Value x() const { return _x; } + Condition cond() const { return _cond; } + bool unordered_is_true() const { return check_flag(UnorderedIsTrueFlag); } + Value y() const { return _y; } + const char *message() const { return _message; } + + // generic + virtual void input_values_do(ValueVisitor* f) { f->visit(&_x); f->visit(&_y); } +}; + +#endif + +LEAF(RangeCheckPredicate, StateSplit) + private: + Value _x; + Condition _cond; + Value _y; + + void check_state(); + + public: + // creation + // unordered_is_true is valid for float/double compares only + RangeCheckPredicate(Value x, Condition cond, bool unordered_is_true, Value y, ValueStack* state) : StateSplit(illegalType) + , _x(x) + , _cond(cond) + , _y(y) + { + ASSERT_VALUES + set_flag(UnorderedIsTrueFlag, unordered_is_true); + assert(x->type()->tag() == y->type()->tag(), "types must match"); + this->set_state(state); + check_state(); + } + + // Always deoptimize + RangeCheckPredicate(ValueStack* state) : StateSplit(illegalType) + { + this->set_state(state); + _x = _y = NULL; + check_state(); + } + + // accessors + Value x() const { return _x; } + Condition cond() const { return _cond; } + bool unordered_is_true() const { return check_flag(UnorderedIsTrueFlag); } + Value y() const { return _y; } + + void always_fail() { _x = _y = NULL; } + + // generic + virtual void input_values_do(ValueVisitor* f) { StateSplit::input_values_do(f); f->visit(&_x); f->visit(&_y); } + HASHING3(RangeCheckPredicate, true, x()->subst(), y()->subst(), cond()) +}; LEAF(If, BlockEnd) private: --- old/src/share/vm/c1/c1_InstructionPrinter.cpp 2013-02-20 17:40:47.182026331 +0100 +++ new/src/share/vm/c1/c1_InstructionPrinter.cpp 2013-02-20 17:40:47.011023103 +0100 @@ -57,6 +57,8 @@ case If::leq: return "<="; case If::gtr: return ">"; case If::geq: return ">="; + case If::aeq: return "aeq"; + case If::beq: return "beq"; } ShouldNotReachHere(); return NULL; @@ -181,6 +183,11 @@ output()->put('['); print_value(indexed->index()); output()->put(']'); + if (indexed->length() != NULL) { + output()->put('('); + print_value(indexed->length()); + output()->put(')'); + } } @@ -373,6 +380,7 @@ void InstructionPrinter::do_LoadField(LoadField* x) { print_field(x); output()->print(" (%c)", type2char(x->field()->type()->basic_type())); + output()->print(" %s", x->field()->name()->as_utf8()); } @@ -381,6 +389,7 @@ output()->print(" := "); print_value(x->value()); output()->print(" (%c)", type2char(x->field()->type()->basic_type())); + output()->print(" %s", x->field()->name()->as_utf8()); } @@ -393,6 +402,9 @@ void InstructionPrinter::do_LoadIndexed(LoadIndexed* x) { print_indexed(x); output()->print(" (%c)", type2char(x->elt_type())); + if (x->check_flag(Instruction::NeedsRangeCheckFlag)) { + output()->print(" [rc]"); + } } @@ -401,6 +413,9 @@ output()->print(" := "); print_value(x->value()); output()->print(" (%c)", type2char(x->elt_type())); + if (x->check_flag(Instruction::NeedsRangeCheckFlag)) { + output()->print(" [rc]"); + } } void InstructionPrinter::do_NegateOp(NegateOp* x) { @@ -843,6 +858,25 @@ output()->put(')'); } +void InstructionPrinter::do_RangeCheckPredicate(RangeCheckPredicate* x) { + + if (x->x() != NULL && x->y() != NULL) { + output()->print("if "); + print_value(x->x()); + output()->print(" %s ", cond_name(x->cond())); + print_value(x->y()); + output()->print(" then deoptimize!"); + } else { + output()->print("always deoptimize!"); + } +} + +void InstructionPrinter::do_Assert(Assert* x) { + output()->print("assert "); + print_value(x->x()); + output()->print(" %s ", cond_name(x->cond())); + print_value(x->y()); +} void InstructionPrinter::do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) { print_unsafe_object_op(x, "UnsafePrefetchWrite"); --- old/src/share/vm/c1/c1_InstructionPrinter.hpp 2013-02-20 17:40:48.528378232 +0100 +++ new/src/share/vm/c1/c1_InstructionPrinter.hpp 2013-02-20 17:40:48.351690818 +0100 @@ -135,6 +135,8 @@ virtual void do_ProfileInvoke (ProfileInvoke* x); virtual void do_RuntimeCall (RuntimeCall* x); virtual void do_MemBar (MemBar* x); + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x); + virtual void do_Assert (Assert* x); }; #endif // PRODUCT --- old/src/share/vm/c1/c1_LIR.cpp 2013-02-20 17:40:49.789649671 +0100 +++ new/src/share/vm/c1/c1_LIR.cpp 2013-02-20 17:40:49.627895853 +0100 @@ -633,6 +633,7 @@ case lir_ushr: case lir_xadd: case lir_xchg: + case lir_assert: { assert(op->as_Op2() != NULL, "must be"); LIR_Op2* op2 = (LIR_Op2*)op; @@ -1112,6 +1113,11 @@ } } +#ifndef PRODUCT +void LIR_OpAssert::emit_code(LIR_Assembler* masm) { + masm->emit_assert(this); +} +#endif void LIR_OpDelay::emit_code(LIR_Assembler* masm) { masm->emit_delay(this); @@ -1771,6 +1777,8 @@ case lir_cas_int: s = "cas_int"; break; // LIR_OpProfileCall case lir_profile_call: s = "profile_call"; break; + // LIR_OpAssert + case lir_assert: s = "assert"; break; case lir_none: ShouldNotReachHere();break; default: s = "illegal_op"; break; } @@ -2017,6 +2025,13 @@ out->print("[lbl:0x%x]", stub()->entry()); } +void LIR_OpAssert::print_instr(outputStream* out) const { + print_condition(out, condition()); out->print(" "); + in_opr1()->print(out); out->print(" "); + in_opr2()->print(out); out->print(", \""); + out->print(msg()); out->print("\""); +} + void LIR_OpDelay::print_instr(outputStream* out) const { _op->print_on(out); --- old/src/share/vm/c1/c1_LIR.hpp 2013-02-20 17:40:51.145440802 +0100 +++ new/src/share/vm/c1/c1_LIR.hpp 2013-02-20 17:40:50.976329383 +0100 @@ -881,6 +881,7 @@ class LIR_OpTypeCheck; class LIR_OpCompareAndSwap; class LIR_OpProfileCall; +class LIR_OpAssert; // LIR operation codes @@ -1000,6 +1001,9 @@ , begin_opMDOProfile , lir_profile_call , end_opMDOProfile + , begin_opAssert + , lir_assert + , end_opAssert }; @@ -1135,6 +1139,7 @@ virtual LIR_OpTypeCheck* as_OpTypeCheck() { return NULL; } virtual LIR_OpCompareAndSwap* as_OpCompareAndSwap() { return NULL; } virtual LIR_OpProfileCall* as_OpProfileCall() { return NULL; } + virtual LIR_OpAssert* as_OpAssert() { return NULL; } virtual void verify() const {} }; @@ -1623,7 +1628,7 @@ , _tmp3(LIR_OprFact::illegalOpr) , _tmp4(LIR_OprFact::illegalOpr) , _tmp5(LIR_OprFact::illegalOpr) { - assert(code == lir_cmp, "code check"); + assert(code == lir_cmp || code == lir_assert, "code check"); } LIR_Op2(LIR_Code code, LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, LIR_Opr result, BasicType type) @@ -1683,7 +1688,7 @@ LIR_Opr tmp4_opr() const { return _tmp4; } LIR_Opr tmp5_opr() const { return _tmp5; } LIR_Condition condition() const { - assert(code() == lir_cmp || code() == lir_cmove, "only valid for cmp and cmove"); return _condition; + assert(code() == lir_cmp || code() == lir_cmove || code() == lir_assert, "only valid for cmp and cmove and assert"); return _condition; } void set_condition(LIR_Condition condition) { assert(code() == lir_cmp || code() == lir_cmove, "only valid for cmp and cmove"); _condition = condition; @@ -1823,6 +1828,30 @@ CodeEmitInfo* call_info() const { return info(); } }; +#ifndef PRODUCT +// LIR_OpAssert +class LIR_OpAssert : public LIR_Op2 { + friend class LIR_OpVisitState; + + private: + const char* _msg; + bool _halt; + + public: + LIR_OpAssert(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, const char* msg, bool halt) + : LIR_Op2(lir_assert, condition, opr1, opr2) + , _halt(halt) + , _msg(msg) { + } + + const char* msg() const { return _msg; } + bool halt() const { return _halt; } + + virtual void emit_code(LIR_Assembler* masm); + virtual LIR_OpAssert* as_OpAssert() { return this; } + virtual void print_instr(outputStream* out) const PRODUCT_RETURN; +}; +#endif // LIR_OpCompareAndSwap class LIR_OpCompareAndSwap : public LIR_Op { @@ -2196,6 +2225,9 @@ void xadd(LIR_Opr src, LIR_Opr add, LIR_Opr res, LIR_Opr tmp) { append(new LIR_Op2(lir_xadd, src, add, res, tmp)); } void xchg(LIR_Opr src, LIR_Opr set, LIR_Opr res, LIR_Opr tmp) { append(new LIR_Op2(lir_xchg, src, set, res, tmp)); } +#ifndef PRODUCT + void lir_assert(LIR_Condition condition, LIR_Opr opr1, LIR_Opr opr2, const char* msg, bool halt) { append(new LIR_OpAssert(condition, opr1, opr2, msg, halt)); } +#endif }; void print_LIR(BlockList* blocks); --- old/src/share/vm/c1/c1_LIRAssembler.hpp 2013-02-20 17:40:52.495605537 +0100 +++ new/src/share/vm/c1/c1_LIRAssembler.hpp 2013-02-20 17:40:52.327531467 +0100 @@ -210,6 +210,7 @@ void arith_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest, CodeEmitInfo* info, bool pop_fpu_stack); void arithmetic_idiv(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr temp, LIR_Opr result, CodeEmitInfo* info); void intrinsic_op(LIR_Code code, LIR_Opr value, LIR_Opr unused, LIR_Opr dest, LIR_Op* op); + NOT_PRODUCT(void emit_assert(LIR_OpAssert* op);) void logic_op(LIR_Code code, LIR_Opr left, LIR_Opr right, LIR_Opr dest); --- old/src/share/vm/c1/c1_LIRGenerator.cpp 2013-02-20 17:40:53.893214583 +0100 +++ new/src/share/vm/c1/c1_LIRGenerator.cpp 2013-02-20 17:40:53.701733229 +0100 @@ -403,6 +403,10 @@ CodeEmitInfo* LIRGenerator::state_for(Instruction* x, ValueStack* state, bool ignore_xhandler) { assert(state != NULL, "state must be defined"); +#ifndef PRODUCT + state->verify(); +#endif + ValueStack* s = state; for_each_state(s) { if (s->kind() == ValueStack::EmptyExceptionState) { @@ -453,7 +457,7 @@ } } - return new CodeEmitInfo(state, ignore_xhandler ? NULL : x->exception_handlers()); + return new CodeEmitInfo(state, ignore_xhandler ? NULL : x->exception_handlers(), x->check_flag(Instruction::DeoptimizeOnException)); } @@ -1791,12 +1795,19 @@ x->is_static() ? "static" : "field", x->printable_bci()); } #endif - + + bool stress_deopt = StressLoopInvariantCodeMotion && info && info->deoptimize_on_exception(); if (x->needs_null_check() && (needs_patching || - MacroAssembler::needs_explicit_null_check(x->offset()))) { + MacroAssembler::needs_explicit_null_check(x->offset()) || + stress_deopt)) { + LIR_Opr obj = object.result(); + if (stress_deopt) { + obj = new_register(T_OBJECT); + __ move(LIR_OprFact::oopConst(NULL), obj); + } // emit an explicit null check because the offset is too large - __ null_check(object.result(), new CodeEmitInfo(info)); + __ null_check(obj, new CodeEmitInfo(info)); } LIR_Opr reg = rlock_result(x, field_type); @@ -1861,6 +1872,8 @@ void LIRGenerator::do_ArrayLength(ArrayLength* x) { + if (x->use_count() == 0 && !x->can_trap()) return; + LIRItem array(x->array(), this); array.load_item(); LIR_Opr reg = rlock_result(x); @@ -1873,6 +1886,11 @@ } else { info = state_for(nc); } + if (StressLoopInvariantCodeMotion && info->deoptimize_on_exception()) { + LIR_Opr obj = new_register(T_OBJECT); + __ move(LIR_OprFact::oopConst(NULL), obj); + __ null_check(obj, new CodeEmitInfo(info)); + } } __ load(new LIR_Address(array.result(), arrayOopDesc::length_offset_in_bytes(), T_INT), reg, info, lir_patch_none); } @@ -1883,14 +1901,11 @@ LIRItem array(x->array(), this); LIRItem index(x->index(), this); LIRItem length(this); - bool needs_range_check = true; + bool needs_range_check = x->compute_needs_range_check(); - if (use_length) { - needs_range_check = x->compute_needs_range_check(); - if (needs_range_check) { - length.set_instruction(x->length()); - length.load_item(); - } + if (use_length && needs_range_check) { + length.set_instruction(x->length()); + length.load_item(); } array.load_item(); @@ -1910,13 +1925,20 @@ } else { null_check_info = range_check_info; } + if (StressLoopInvariantCodeMotion && null_check_info->deoptimize_on_exception()) { + LIR_Opr obj = new_register(T_OBJECT); + __ move(LIR_OprFact::oopConst(NULL), obj); + __ null_check(obj, new CodeEmitInfo(null_check_info)); + } } // emit array address setup early so it schedules better LIR_Address* array_addr = emit_array_address(array.result(), index.result(), x->elt_type(), false); if (GenerateRangeChecks && needs_range_check) { - if (use_length) { + if (StressLoopInvariantCodeMotion && range_check_info->deoptimize_on_exception()) { + __ branch(lir_cond_always, T_ILLEGAL, new RangeCheckStub(range_check_info, index.result())); + } else if (use_length) { // TODO: use a (modified) version of array_range_check that does not require a // constant length to be loaded to a register __ cmp(lir_cond_belowEqual, length.result(), index.result()); @@ -2634,7 +2656,7 @@ LIR_Opr lock = new_register(T_INT); __ load_stack_address_monitor(0, lock); - CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL); + CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL, x->check_flag(Instruction::DeoptimizeOnException)); CodeStub* slow_path = new MonitorEnterStub(obj, lock, info); // receiver is guaranteed non-NULL so don't need CodeEmitInfo @@ -2644,7 +2666,7 @@ // increment invocation counters if needed if (!method()->is_accessor()) { // Accessors do not have MDOs, so no counting. - CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL); + CodeEmitInfo* info = new CodeEmitInfo(scope()->start()->state()->copy(ValueStack::StateBefore, SynchronizationEntryBCI), NULL, false); increment_invocation_counter(info); } @@ -3102,6 +3124,96 @@ } } +void LIRGenerator::do_Assert(Assert *x) { +#ifndef PRODUCT + ValueTag tag = x->x()->type()->tag(); + If::Condition cond = x->cond(); + + LIRItem xitem(x->x(), this); + LIRItem yitem(x->y(), this); + LIRItem* xin = &xitem; + LIRItem* yin = &yitem; + + assert(tag == intTag, "Only integer assertions are valid!"); + + xin->load_item(); + yin->dont_load_item(); + + set_no_result(x); + + LIR_Opr left = xin->result(); + LIR_Opr right = yin->result(); + + __ lir_assert(lir_cond(x->cond()), left, right, x->message(), true); + +#endif +} + + +void LIRGenerator::do_RangeCheckPredicate(RangeCheckPredicate *x) { + + + Instruction *a = x->x(); + Instruction *b = x->y(); + if (!a || StressRangeCheckElimination) { + assert(!b || StressRangeCheckElimination, "B must also be null"); + + CodeEmitInfo *info = state_for(x, x->state()); + CodeStub* stub = new PredicateFailedStub(info); + + __ jump(stub); + } else if (a->type()->as_IntConstant() && b->type()->as_IntConstant()) { + int a_int = a->type()->as_IntConstant()->value(); + int b_int = b->type()->as_IntConstant()->value(); + + bool ok = false; + + switch(x->cond()) { + case Instruction::eql: ok = (a_int == b_int); break; + case Instruction::neq: ok = (a_int != b_int); break; + case Instruction::lss: ok = (a_int < b_int); break; + case Instruction::leq: ok = (a_int <= b_int); break; + case Instruction::gtr: ok = (a_int > b_int); break; + case Instruction::geq: ok = (a_int >= b_int); break; + case Instruction::aeq: ok = ((unsigned int)a_int >= (unsigned int)b_int); break; + case Instruction::beq: ok = ((unsigned int)a_int <= (unsigned int)b_int); break; + default: ShouldNotReachHere(); + } + + if (ok) { + + CodeEmitInfo *info = state_for(x, x->state()); + CodeStub* stub = new PredicateFailedStub(info); + + __ jump(stub); + } + } else { + + ValueTag tag = x->x()->type()->tag(); + If::Condition cond = x->cond(); + LIRItem xitem(x->x(), this); + LIRItem yitem(x->y(), this); + LIRItem* xin = &xitem; + LIRItem* yin = &yitem; + + assert(tag == intTag, "Only integer deoptimizations are valid!"); + + xin->load_item(); + yin->dont_load_item(); + set_no_result(x); + + LIR_Opr left = xin->result(); + LIR_Opr right = yin->result(); + + CodeEmitInfo *info = state_for(x, x->state()); + CodeStub* stub = new PredicateFailedStub(info); + + __ cmp(lir_cond(cond), left, right); + __ branch(lir_cond(cond), right->type(), stub); + } +} + + LIR_Opr LIRGenerator::call_runtime(Value arg1, address entry, ValueType* result_type, CodeEmitInfo* info) { LIRItemList args(1); LIRItem value(arg1, this); --- old/src/share/vm/c1/c1_LIRGenerator.hpp 2013-02-20 17:40:55.373483063 +0100 +++ new/src/share/vm/c1/c1_LIRGenerator.hpp 2013-02-20 17:40:55.204146640 +0100 @@ -412,6 +412,8 @@ case If::leq: l = lir_cond_lessEqual; break; case If::geq: l = lir_cond_greaterEqual; break; case If::gtr: l = lir_cond_greater; break; + case If::aeq: l = lir_cond_aboveEqual; break; + case If::beq: l = lir_cond_belowEqual; break; }; return l; } @@ -534,6 +536,8 @@ virtual void do_ProfileInvoke (ProfileInvoke* x); virtual void do_RuntimeCall (RuntimeCall* x); virtual void do_MemBar (MemBar* x); + virtual void do_RangeCheckPredicate(RangeCheckPredicate* x); + virtual void do_Assert (Assert* x); }; --- old/src/share/vm/c1/c1_LinearScan.cpp 2013-02-20 17:40:56.765672042 +0100 +++ new/src/share/vm/c1/c1_LinearScan.cpp 2013-02-20 17:40:56.573538900 +0100 @@ -6231,26 +6231,29 @@ assert(prev_op->as_OpBranch() != NULL, "branch must be of type LIR_OpBranch"); LIR_OpBranch* prev_branch = (LIR_OpBranch*)prev_op; - LIR_Op2* prev_cmp = NULL; + if (prev_branch->stub() == NULL) { - for(int j = instructions->length() - 3; j >= 0 && prev_cmp == NULL; j--) { - prev_op = instructions->at(j); - if(prev_op->code() == lir_cmp) { - assert(prev_op->as_Op2() != NULL, "branch must be of type LIR_Op2"); - prev_cmp = (LIR_Op2*)prev_op; - assert(prev_branch->cond() == prev_cmp->condition(), "should be the same"); + LIR_Op2* prev_cmp = NULL; + + for(int j = instructions->length() - 3; j >= 0 && prev_cmp == NULL; j--) { + prev_op = instructions->at(j); + if (prev_op->code() == lir_cmp) { + assert(prev_op->as_Op2() != NULL, "branch must be of type LIR_Op2"); + prev_cmp = (LIR_Op2*)prev_op; + assert(prev_branch->cond() == prev_cmp->condition(), "should be the same"); + } + } + assert(prev_cmp != NULL, "should have found comp instruction for branch"); + if (prev_branch->block() == code->at(i + 1) && prev_branch->info() == NULL) { + + TRACE_LINEAR_SCAN(3, tty->print_cr("Negating conditional branch and deleting unconditional branch at end of block B%d", block->block_id())); + + // eliminate a conditional branch to the immediate successor + prev_branch->change_block(last_branch->block()); + prev_branch->negate_cond(); + prev_cmp->set_condition(prev_branch->cond()); + instructions->truncate(instructions->length() - 1); } - } - assert(prev_cmp != NULL, "should have found comp instruction for branch"); - if (prev_branch->block() == code->at(i + 1) && prev_branch->info() == NULL) { - - TRACE_LINEAR_SCAN(3, tty->print_cr("Negating conditional branch and deleting unconditional branch at end of block B%d", block->block_id())); - - // eliminate a conditional branch to the immediate successor - prev_branch->change_block(last_branch->block()); - prev_branch->negate_cond(); - prev_cmp->set_condition(prev_branch->cond()); - instructions->truncate(instructions->length() - 1); } } } --- old/src/share/vm/c1/c1_Optimizer.cpp 2013-02-20 17:40:58.795295982 +0100 +++ new/src/share/vm/c1/c1_Optimizer.cpp 2013-02-20 17:40:58.618717297 +0100 @@ -178,7 +178,7 @@ // 2) substitute conditional expression // with an IfOp followed by a Goto // cut if_ away and get node before - Instruction* cur_end = if_->prev(block); + Instruction* cur_end = if_->prev(); // append constants of true- and false-block if necessary // clone constants because original block must not be destroyed @@ -202,7 +202,7 @@ } // append Goto to successor - ValueStack* state_before = if_->is_safepoint() ? if_->state_before() : NULL; + ValueStack* state_before = if_->state_before(); Goto* goto_ = new Goto(sux, state_before, if_->is_safepoint() || t_goto->is_safepoint() || f_goto->is_safepoint()); // prepare state for Goto @@ -367,10 +367,11 @@ #endif // find instruction before end & append first instruction of sux block - Instruction* prev = end->prev(block); + Instruction* prev = end->prev(); Instruction* next = sux->next(); assert(prev->as_BlockEnd() == NULL, "must not be a BlockEnd"); prev->set_next(next); + prev->fixup_block_pointers(); sux->disconnect_from_graph(); block->set_end(sux->end()); // add exception handlers of deleted block, if any @@ -533,6 +534,8 @@ void do_ProfileInvoke (ProfileInvoke* x); void do_RuntimeCall (RuntimeCall* x); void do_MemBar (MemBar* x); + void do_RangeCheckPredicate(RangeCheckPredicate* x); + void do_Assert (Assert* x); }; @@ -714,6 +717,8 @@ void NullCheckVisitor::do_ProfileInvoke (ProfileInvoke* x) {} void NullCheckVisitor::do_RuntimeCall (RuntimeCall* x) {} void NullCheckVisitor::do_MemBar (MemBar* x) {} +void NullCheckVisitor::do_RangeCheckPredicate(RangeCheckPredicate* x) {} +void NullCheckVisitor::do_Assert (Assert* x) {} void NullCheckEliminator::visit(Value* p) { --- old/src/share/vm/c1/c1_Runtime1.cpp 2013-02-20 17:41:00.210842783 +0100 +++ new/src/share/vm/c1/c1_Runtime1.cpp 2013-02-20 17:41:00.038187956 +0100 @@ -1330,6 +1330,50 @@ return (k != NULL && obj != NULL && obj->is_a(k)) ? 1 : 0; JRT_END +JRT_ENTRY(void, Runtime1::predicate_failed_trap(JavaThread* thread)) + ResourceMark rm; + + assert(!TieredCompilation, "incompatible with tiered compilation"); + + RegisterMap reg_map(thread, false); + frame runtime_frame = thread->last_frame(); + frame caller_frame = runtime_frame.sender(®_map); + + nmethod* nm = CodeCache::find_nmethod(caller_frame.pc()); + assert (nm != NULL, "no more nmethod?"); + nm->make_not_entrant(); + + methodHandle m(nm->method()); + MethodData* mdo = m->method_data(); + + if (mdo == NULL && !HAS_PENDING_EXCEPTION) { + // Build an MDO. Ignore errors like OutOfMemory; + // that simply means we won't have an MDO to update. + Method::build_interpreter_method_data(m, THREAD); + if (HAS_PENDING_EXCEPTION) { + assert((PENDING_EXCEPTION->is_a(SystemDictionary::OutOfMemoryError_klass())), "we expect only an OOM error here"); + CLEAR_PENDING_EXCEPTION; + } + mdo = m->method_data(); + } + + if (mdo != NULL) { + mdo->inc_trap_count(Deoptimization::Reason_none); + } + + if (TracePredicateFailedTraps) { + stringStream ss1, ss2; + vframeStream vfst(thread); + methodHandle inlinee = methodHandle(vfst.method()); + inlinee->print_short_name(&ss1); + m->print_short_name(&ss2); + tty->print_cr("Predicate failed trap in method %s at bci %d inlined in %s at pc %x", ss1.as_string(), vfst.bci(), ss2.as_string(), caller_frame.pc()); + } + + + Deoptimization::deoptimize_frame(thread, caller_frame.id()); + +JRT_END #ifndef PRODUCT void Runtime1::print_statistics() { --- old/src/share/vm/c1/c1_Runtime1.hpp 2013-02-20 17:41:01.560646549 +0100 +++ new/src/share/vm/c1/c1_Runtime1.hpp 2013-02-20 17:41:01.401028382 +0100 @@ -71,6 +71,7 @@ stub(g1_post_barrier_slow) \ stub(fpu2long_stub) \ stub(counter_overflow) \ + stub(predicate_failed_trap) \ last_entry(number_of_ids) #define DECLARE_STUB_ID(x) x ## _id , @@ -190,6 +191,8 @@ static void oop_arraycopy(HeapWord* src, HeapWord* dst, int length); static int is_instance_of(oopDesc* mirror, oopDesc* obj); + static void predicate_failed_trap(JavaThread* thread); + static void print_statistics() PRODUCT_RETURN; }; --- old/src/share/vm/c1/c1_ValueMap.cpp 2013-02-20 17:41:02.817711944 +0100 +++ new/src/share/vm/c1/c1_ValueMap.cpp 2013-02-20 17:41:02.658322461 +0100 @@ -26,9 +26,9 @@ #include "c1/c1_Canonicalizer.hpp" #include "c1/c1_IR.hpp" #include "c1/c1_ValueMap.hpp" +#include "c1/c1_ValueStack.hpp" #include "utilities/bitMap.inline.hpp" - #ifndef PRODUCT int ValueMap::_number_of_finds = 0; @@ -192,10 +192,6 @@ && lf->field()->holder() == field->holder() \ && (all_offsets || lf->field()->offset() == field->offset()); -#define MUST_KILL_EXCEPTION(must_kill, entry, value) \ - assert(entry->nesting() < nesting(), "must not find bigger nesting than current"); \ - bool must_kill = (entry->nesting() == nesting() - 1); - void ValueMap::kill_memory() { GENERIC_KILL_VALUE(MUST_KILL_MEMORY); @@ -209,11 +205,6 @@ GENERIC_KILL_VALUE(MUST_KILL_FIELD); } -void ValueMap::kill_exception() { - GENERIC_KILL_VALUE(MUST_KILL_EXCEPTION); -} - - void ValueMap::kill_map(ValueMap* map) { assert(is_global_value_numbering(), "only for global value numbering"); _killed_values.set_union(&map->_killed_values); @@ -274,6 +265,8 @@ GlobalValueNumbering* _gvn; BlockList _loop_blocks; bool _too_complicated_loop; + bool _has_field_store[T_ARRAY + 1]; + bool _has_indexed_store[T_ARRAY + 1]; // simplified access to methods of GlobalValueNumbering ValueMap* current_map() { return _gvn->current_map(); } @@ -281,8 +274,16 @@ // implementation for abstract methods of ValueNumberingVisitor void kill_memory() { _too_complicated_loop = true; } - void kill_field(ciField* field, bool all_offsets) { current_map()->kill_field(field, all_offsets); }; - void kill_array(ValueType* type) { current_map()->kill_array(type); }; + void kill_field(ciField* field, bool all_offsets) { + current_map()->kill_field(field, all_offsets); + assert(field->type()->basic_type() >= 0 && field->type()->basic_type() <= T_ARRAY, "Invalid type"); + _has_field_store[field->type()->basic_type()] = true; + } + void kill_array(ValueType* type) { + current_map()->kill_array(type); + BasicType basic_type = as_BasicType(type); assert(basic_type >= 0 && basic_type <= T_ARRAY, "Invalid type"); + _has_indexed_store[basic_type] = true; + } public: ShortLoopOptimizer(GlobalValueNumbering* gvn) @@ -290,11 +291,153 @@ , _loop_blocks(ValueMapMaxLoopSize) , _too_complicated_loop(false) { + for (int i=0; i<= T_ARRAY; i++){ + _has_field_store[i] = false; + _has_indexed_store[i] = false; + } + } + + bool has_field_store(BasicType type) { + assert(type >= 0 && type <= T_ARRAY, "Invalid type"); + return _has_field_store[type]; + } + + bool has_indexed_store(BasicType type) { + assert(type >= 0 && type <= T_ARRAY, "Invalid type"); + return _has_indexed_store[type]; } bool process(BlockBegin* loop_header); }; +class LoopInvariantCodeMotion : public StackObj { + private: + GlobalValueNumbering* _gvn; + ShortLoopOptimizer* _short_loop_optimizer; + Instruction* _insertion_point; + ValueStack * _state; + + void set_invariant(Value v) const { _gvn->set_processed(v); } + bool is_invariant(Value v) const { return _gvn->is_processed(v); } + + void process_block(BlockBegin* block); + + public: + LoopInvariantCodeMotion(ShortLoopOptimizer *slo, GlobalValueNumbering* gvn, BlockBegin* loop_header, BlockList* loop_blocks); +}; + +LoopInvariantCodeMotion::LoopInvariantCodeMotion(ShortLoopOptimizer *slo, GlobalValueNumbering* gvn, BlockBegin* loop_header, BlockList* loop_blocks) + : _gvn(gvn), _short_loop_optimizer(slo) +{ + + TRACE_VALUE_NUMBERING(tty->print_cr("using loop invariant code motion loop_header = %d", loop_header->block_id())); + TRACE_VALUE_NUMBERING(tty->print_cr("** loop invariant code motion for short loop B%d", loop_header->block_id())); + + BlockBegin* insertion_block = loop_header->dominator(); + if (insertion_block->number_of_preds() == 0) { + return; // only the entry block does not have a predecessor + } + + assert(insertion_block->end()->as_Base() == NULL, "cannot insert into entry block"); + _insertion_point = insertion_block->end()->prev(); + + BlockEnd *block_end = insertion_block->end(); + _state = block_end->state_before(); + + if (!_state) { + // If, TableSwitch and LookupSwitch always have state_before when + // loop invariant code motion happens.. + assert(block_end->as_Goto(), "Block has to be goto"); + _state = block_end->state(); + } + + // the loop_blocks are filled by going backward from the loop header, so this processing order is best + assert(loop_blocks->at(0) == loop_header, "loop header must be first loop block"); + process_block(loop_header); + for (int i = loop_blocks->length() - 1; i >= 1; i--) { + process_block(loop_blocks->at(i)); + } +} + +void LoopInvariantCodeMotion::process_block(BlockBegin* block) { + + TRACE_VALUE_NUMBERING(tty->print_cr("processing block B%d", block->block_id())); + + Instruction* prev = block; + Instruction* cur = block->next(); + + while (cur != NULL) { + + // determine if cur instruction is loop invariant + // only selected instruction types are processed here + bool cur_invariant = false; + + if (cur->as_Constant() != NULL) { + + cur_invariant = !cur->can_trap(); + + } else if (cur->as_ArithmeticOp() != NULL || cur->as_LogicOp() != NULL || cur->as_ShiftOp() != NULL) { + + assert(cur->as_Op2() != NULL, "must be Op2"); + Op2* op2 = (Op2*)cur; + cur_invariant = !op2->can_trap() && is_invariant(op2->x()) && is_invariant(op2->y()); + + } else if (cur->as_LoadField() != NULL) { + + LoadField* lf = (LoadField*)cur; + + // deoptimizes on NullPointerException + cur_invariant = !lf->needs_patching() && !lf->field()->is_volatile() && !_short_loop_optimizer->has_field_store(lf->field()->type()->basic_type()) && is_invariant(lf->obj()); + } else if (cur->as_ArrayLength() != NULL) { + + ArrayLength *length = cur->as_ArrayLength(); + cur_invariant = is_invariant(length->array()); + } else if (cur->as_LoadIndexed() != NULL) { + + LoadIndexed *li = (LoadIndexed *)cur->as_LoadIndexed(); + cur_invariant = (!_short_loop_optimizer->has_indexed_store(li->elt_type())) && (is_invariant(li->array()) && is_invariant(li->index())); + } + + if (cur_invariant) { + + // perform value numbering and mark instruction as loop-invariant + _gvn->substitute(cur); + + if (cur->as_Constant() == NULL) { + // ensure that code for non-constant instructions is always generated + cur->pin(); + } + + // remove cur instruction from loop block and append it to block before loop + Instruction* next = cur->next(); + + Instruction* in = _insertion_point->next(); + _insertion_point = _insertion_point->set_next(cur); + cur->set_next(in); + + // Deoptimize on exception + cur->set_flag(Instruction::DeoptimizeOnException, true); + + // Clear exception handlers + cur->set_exception_handlers(NULL); + + TRACE_VALUE_NUMBERING(tty->print_cr("Instruction %c%d is loop invariant", cur->type()->tchar(), cur->id())); + + if (cur->state_before() != NULL) { + cur->set_state_before(_state->copy()); + } + if (cur->exception_state() != NULL) { + cur->set_exception_state(_state->copy()); + } + + cur = prev->set_next(next); + + } else { + prev = cur; + cur = cur->next(); + } + } +} bool ShortLoopOptimizer::process(BlockBegin* loop_header) { TRACE_VALUE_NUMBERING(tty->print_cr("** loop header block")); @@ -316,6 +459,10 @@ for (int j = block->number_of_preds() - 1; j >= 0; j--) { BlockBegin* pred = block->pred_at(j); + if (pred->is_set(BlockBegin::osr_entry_flag)) { + return false; + } + ValueMap* pred_map = value_map_of(pred); if (pred_map != NULL) { current_map()->kill_map(pred_map); @@ -336,6 +483,12 @@ } } + bool optimistic = this->_gvn->compilation()->is_optimistic(); + + if (UseLoopInvariantCodeMotion && optimistic) { + LoopInvariantCodeMotion code_motion(this, _gvn, loop_header, &_loop_blocks); + } + TRACE_VALUE_NUMBERING(tty->print_cr("** loop successfully optimized")); return true; } @@ -344,11 +497,11 @@ GlobalValueNumbering::GlobalValueNumbering(IR* ir) : _current_map(NULL) , _value_maps(ir->linear_scan_order()->length(), NULL) + , _compilation(ir->compilation()) { TRACE_VALUE_NUMBERING(tty->print_cr("****** start of global value numbering")); ShortLoopOptimizer short_loop_optimizer(this); - int subst_count = 0; BlockList* blocks = ir->linear_scan_order(); int num_blocks = blocks->length(); @@ -357,6 +510,12 @@ assert(start_block == ir->start() && start_block->number_of_preds() == 0 && start_block->dominator() == NULL, "must be start block"); assert(start_block->next()->as_Base() != NULL && start_block->next()->next() == NULL, "start block must not have instructions"); + // method parameters are not linked in instructions list, so process them separateley + for_each_state_value(start_block->state(), value, + assert(value->as_Local() != NULL, "only method parameters allowed"); + set_processed(value); + ); + // initial, empty value map with nesting 0 set_value_map_of(start_block, new ValueMap()); @@ -374,7 +533,7 @@ // create new value map with increased nesting _current_map = new ValueMap(value_map_of(dominator)); - if (num_preds == 1) { + if (num_preds == 1 && !block->is_set(BlockBegin::exception_entry_flag)) { assert(dominator == block->pred_at(0), "dominator must be equal to predecessor"); // nothing to do here @@ -403,36 +562,41 @@ } } - if (block->is_set(BlockBegin::exception_entry_flag)) { - current_map()->kill_exception(); - } + // phi functions are not linked in instructions list, so process them separateley + for_each_phi_fun(block, phi, + set_processed(phi); + ); TRACE_VALUE_NUMBERING(tty->print("value map before processing block: "); current_map()->print()); // visit all instructions of this block for (Value instr = block->next(); instr != NULL; instr = instr->next()) { - assert(!instr->has_subst(), "substitution already set"); - // check if instruction kills any values instr->visit(this); - - if (instr->hash() != 0) { - Value f = current_map()->find_insert(instr); - if (f != instr) { - assert(!f->has_subst(), "can't have a substitution"); - instr->set_subst(f); - subst_count++; - } - } + // perform actual value numbering + substitute(instr); } // remember value map for successors set_value_map_of(block, current_map()); } - if (subst_count != 0) { + if (_has_substitutions) { SubstitutionResolver resolver(ir); } TRACE_VALUE_NUMBERING(tty->print("****** end of global value numbering. "); ValueMap::print_statistics()); } + +void GlobalValueNumbering::substitute(Instruction* instr) { + assert(!instr->has_subst(), "substitution already set"); + Value subst = current_map()->find_insert(instr); + if (subst != instr) { + assert(!subst->has_subst(), "can't have a substitution"); + + TRACE_VALUE_NUMBERING(tty->print_cr("substitution for %d set to %d", instr->id(), subst->id())); + instr->set_subst(subst); + _has_substitutions = true; + } + set_processed(instr); +} --- old/src/share/vm/c1/c1_ValueMap.hpp 2013-02-20 17:41:04.140537413 +0100 +++ new/src/share/vm/c1/c1_ValueMap.hpp 2013-02-20 17:41:03.988832115 +0100 @@ -206,6 +206,8 @@ void do_ProfileInvoke (ProfileInvoke* x) { /* nothing to do */ }; void do_RuntimeCall (RuntimeCall* x) { /* nothing to do */ }; void do_MemBar (MemBar* x) { /* nothing to do */ }; + void do_RangeCheckPredicate(RangeCheckPredicate* x) { /* nothing to do */ }; + void do_Assert (Assert* x) { /* nothing to do */ }; }; @@ -225,15 +227,22 @@ class GlobalValueNumbering: public ValueNumberingVisitor { private: + Compilation* _compilation; // compilation data ValueMap* _current_map; // value map of current block ValueMapArray _value_maps; // list of value maps for all blocks + ValueSet _processed_values; // marker for instructions that were already processed + bool _has_substitutions; // set to true when substitutions must be resolved public: // accessors + Compilation* compilation() const { return _compilation; } ValueMap* current_map() { return _current_map; } ValueMap* value_map_of(BlockBegin* block) { return _value_maps.at(block->linear_scan_number()); } void set_value_map_of(BlockBegin* block, ValueMap* map) { assert(value_map_of(block) == NULL, ""); _value_maps.at_put(block->linear_scan_number(), map); } + bool is_processed(Value v) { return _processed_values.contains(v); } + void set_processed(Value v) { _processed_values.put(v); } + // implementation for abstract methods of ValueNumberingVisitor void kill_memory() { current_map()->kill_memory(); } void kill_field(ciField* field, bool all_offsets) { current_map()->kill_field(field, all_offsets); } @@ -241,6 +250,7 @@ // main entry point that performs global value numbering GlobalValueNumbering(IR* ir); + void substitute(Instruction* instr); // substitute instruction if it is contained in current value map }; #endif // SHARE_VM_C1_C1_VALUEMAP_HPP --- old/src/share/vm/c1/c1_globals.hpp 2013-02-20 17:41:06.225742569 +0100 +++ new/src/share/vm/c1/c1_globals.hpp 2013-02-20 17:41:06.034377885 +0100 @@ -119,6 +119,24 @@ develop(bool, UseGlobalValueNumbering, true, \ "Use Global Value Numbering (separate phase)") \ \ + product(bool, UseLoopInvariantCodeMotion, true, \ + "Simple loop invariant code motion for short loops during GVN") \ + \ + develop(bool, TracePredicateFailedTraps, false, \ + "trace runtime traps caused by predicate failure") \ + \ + develop(bool, StressLoopInvariantCodeMotion, false, \ + "stress loop invariant code motion") \ + \ + develop(bool, TraceRangeCheckElimination, false, \ + "Trace Range Check Elimination") \ + \ + develop(bool, AssertRangeCheckElimination, false, \ + "Assert Range Check Elimination") \ + \ + develop(bool, StressRangeCheckElimination, false, \ + "stress Range Check Elimination") \ + \ develop(bool, PrintValueNumbering, false, \ "Print Value Numbering") \ \ --- old/src/share/vm/compiler/compileBroker.cpp 2013-02-20 17:41:07.549774812 +0100 +++ new/src/share/vm/compiler/compileBroker.cpp 2013-02-20 17:41:07.369059467 +0100 @@ -2149,6 +2149,9 @@ compiler(CompLevel_full_optimization)->print_timers(); } tty->cr(); + tty->print_cr(" Total compiled methods : %6d methods", CompileBroker::_total_compile_count); + tty->print_cr(" Standard compilation : %6d methods", CompileBroker::_total_standard_compile_count); + tty->print_cr(" On stack replacement : %6d methods", CompileBroker::_total_osr_compile_count); int tcb = CompileBroker::_sum_osr_bytes_compiled + CompileBroker::_sum_standard_bytes_compiled; tty->print_cr(" Total compiled bytecodes : %6d bytes", tcb); tty->print_cr(" Standard compilation : %6d bytes", CompileBroker::_sum_standard_bytes_compiled); --- old/src/share/vm/oops/instanceKlass.cpp 2013-02-20 17:41:09.041251641 +0100 +++ new/src/share/vm/oops/instanceKlass.cpp 2013-02-20 17:41:08.855596608 +0100 @@ -2179,8 +2179,6 @@ } void InstanceKlass::clean_method_data(BoolObjectClosure* is_alive) { -#ifdef COMPILER2 - // Currently only used by C2. for (int m = 0; m < methods()->length(); m++) { MethodData* mdo = methods()->at(m)->method_data(); if (mdo != NULL) { @@ -2191,15 +2189,6 @@ } } } -#else -#ifdef ASSERT - // Verify that we haven't started to use MDOs for C1. - for (int m = 0; m < methods()->length(); m++) { - MethodData* mdo = methods()->at(m)->method_data(); - assert(mdo == NULL, "Didn't expect C1 to use MDOs"); - } -#endif // ASSERT -#endif // !COMPILER2 } --- old/src/share/vm/oops/methodData.cpp 2013-02-20 17:41:10.477242813 +0100 +++ new/src/share/vm/oops/methodData.cpp 2013-02-20 17:41:10.309562666 +0100 @@ -392,6 +392,9 @@ } int MethodData::bytecode_cell_count(Bytecodes::Code code) { +#if defined(COMPILER1) && !defined(COMPILER2) + return no_profile_data; +#endif switch (code) { case Bytecodes::_checkcast: case Bytecodes::_instanceof: @@ -509,6 +512,9 @@ // the segment in bytes. int MethodData::initialize_data(BytecodeStream* stream, int data_index) { +#if defined(COMPILER1) && !defined(COMPILER2) + return 0; +#endif int cell_count = -1; int tag = DataLayout::no_tag; DataLayout* data_layout = data_layout_at(data_index); --- old/src/share/vm/runtime/globals.hpp 2013-02-20 17:41:11.941036351 +0100 +++ new/src/share/vm/runtime/globals.hpp 2013-02-20 17:41:11.730381930 +0100 @@ -2511,7 +2511,7 @@ "disable locking assertions (for speed)") \ \ product(bool, RangeCheckElimination, true, \ - "Split loop iterations to eliminate range checks") \ + "Eliminate range checks") \ \ develop_pd(bool, UncommonNullCast, \ "track occurrences of null in casts; adjust compiler tactics") \ --- /dev/null 2012-11-23 07:18:02.922988661 +0100 +++ new/src/share/vm/c1/c1_RangeCheckElimination.cpp 2013-02-20 17:41:13.307887046 +0100 @@ -0,0 +1,1502 @@ +/* + * Copyright (c) 2012, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA + * or visit www.oracle.com if you need additional information or have any + * questions. + * + */ + +#include "precompiled.hpp" +#include "c1/c1_ValueStack.hpp" +#include "c1/c1_RangeCheckElimination.hpp" +#include "c1/c1_IR.hpp" +#include "c1/c1_Canonicalizer.hpp" +#include "c1/c1_ValueMap.hpp" +#include "ci/ciMethodData.hpp" +#include "runtime/deoptimization.hpp" + +// Macros for the Trace and the Assertion flag +#ifndef PRODUCT +#define TRACE_RANGE_CHECK_ELIMINATION(code) if (TraceRangeCheckElimination) { code; } +#define ASSERT_RANGE_CHECK_ELIMINATION(code) if (AssertRangeCheckElimination) { code; } +#define TRACE_OR_ASSERT_RANGE_CHECK_ELIMINATION(code) if (TraceRangeCheckElimination || AssertRangeCheckElimination) { code; } +#else +#define TRACE_RANGE_CHECK_ELIMINATION(code) +#define ASSERT_RANGE_CHECK_ELIMINATION(code) +#define TRACE_OR_ASSERT_RANGE_CHECK_ELIMINATION(code) +#endif + +// Entry point for the optimization +void RangeCheckElimination::eliminate(IR *ir) { + bool do_elimination = ir->compilation()->has_access_indexed(); + ASSERT_RANGE_CHECK_ELIMINATION(do_elimination = true); + if (do_elimination) { + RangeCheckEliminator rce(ir); + } +} + +// Constructor +RangeCheckEliminator::RangeCheckEliminator(IR *ir) : + _bounds(Instruction::number_of_instructions(), NULL), + _access_indexed_info(Instruction::number_of_instructions(), NULL) +{ + _visitor.set_range_check_eliminator(this); + _ir = ir; + _number_of_instructions = Instruction::number_of_instructions(); + _optimistic = ir->compilation()->is_optimistic(); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr(""); + tty->print_cr("Range check elimination"); + ir->method()->print_name(tty); + tty->print_cr(""); + ); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("optimistic=%d", (int)_optimistic); + ); + +#ifdef ASSERT + // Verifies several conditions that must be true on the IR-input. Only used for debugging purposes. + TRACE_RANGE_CHECK_ELIMINATION( + tty->print_cr("Verification of IR . . ."); + ); + Verification verification(ir); +#endif + + // Set process block flags + // Optimization so a blocks is only processed if it contains an access indexed instruction or if + // one of its children in the dominator tree contains an access indexed instruction. + set_process_block_flags(ir->start()); + + // Pass over instructions in the dominator tree + TRACE_RANGE_CHECK_ELIMINATION(tty->print_cr("Starting pass over dominator tree . . .")); + calc_bounds(ir->start(), NULL); + + TRACE_RANGE_CHECK_ELIMINATION(tty->print_cr("Finished!")); +} + +// Instruction specific work for some instructions +// Constant +void RangeCheckEliminator::Visitor::do_Constant(Constant *c) { + IntConstant *ic; + if (ic = c->type()->as_IntConstant()) { + int value = ic->value(); + _bound = new Bound(value, NULL, value, NULL); + } +} + +// LogicOp +void RangeCheckEliminator::Visitor::do_LogicOp(LogicOp *lo) { + Constant *c; + if (lo->type()->as_IntType() && lo->op() == Bytecodes::_iand && (lo->x()->as_Constant() || lo->y()->as_Constant())) { + int constant = 0; + if (c = lo->x()->as_Constant()) { + constant = c->type()->as_IntConstant()->value(); + } else { + constant = lo->y()->as_Constant()->type()->as_IntConstant()->value(); + } + if (constant >= 0) { + _bound = new Bound(0, NULL, constant, NULL); + } + } +} + +// Phi +void RangeCheckEliminator::Visitor::do_Phi(Phi *phi) +{ + if (!phi->type()->as_IntType() && !phi->type()->as_ObjectType()) return; + + ArithmeticOp *ao; + Constant *c; + Op2 *op2; + BlockBegin *block = phi->block(); + int op_count = phi->operand_count(); + bool has_upper = true; + bool has_lower = true; + assert(phi, "Phi must not be null"); + Bound *bound = NULL; + + // TODO: support more difficult phis + for (int i=0; ioperand_at(i); + + if (v == phi) continue; + + // Check if instruction is connected with phi itself + if ((op2 = v->as_Op2())) { + Value x = op2->x(); + Value y = op2->y(); + if ((x == phi || y == phi)) { + Value other = x; + if (other == phi) { + other = y; + } + if ((ao = v->as_ArithmeticOp()) && ao->op() == Bytecodes::_iadd) { + assert(ao->op() == Bytecodes::_iadd, "Has to be add!"); + if (ao->type()->as_IntType()) { + if (c = other->as_Constant()) { + assert(c->type()->as_IntConstant(), "Constant has to be of type integer"); + int value = c->type()->as_IntConstant()->value(); + if (value == 1) { + has_upper = false; + } else if (value > 1) { + // Overflow not guaranteed + has_upper = false; + has_lower = false; + } else if (value < 0) { + has_lower = false; + } + continue; + } + } + } + } + } + + // No connection -> new bound + Bound *v_bound = _rce->get_bound(v); + Bound *cur_bound; + int cur_constant = 0; + Value cur_value = v; + + if (v->type()->as_IntConstant()) { + cur_constant = v->type()->as_IntConstant()->value(); + cur_value = NULL; + } + if (!v_bound->has_upper() || !v_bound->has_lower()) { + cur_bound = new Bound(cur_constant, cur_value, cur_constant, cur_value); + } else { + cur_bound = v_bound; + } + if (cur_bound) { + if (!bound) { + bound = cur_bound->copy(); + } else { + bound->or_op(cur_bound); + } + } else { + // No bound! + bound = NULL; + break; + } + } + + if (bound) { + if (!has_upper) { + bound->remove_upper(); + } + if (!has_lower) { + bound->remove_lower(); + } + _bound = bound; + } else { + _bound = new Bound(); + } +} + + +// ArithmeticOp +void RangeCheckEliminator::Visitor::do_ArithmeticOp(ArithmeticOp *ao) { + Value x = ao->x(); + Value y = ao->y(); + + if (ao->op() == Bytecodes::_irem) { + Bound* x_bound = _rce->get_bound(x); + Bound* y_bound = _rce->get_bound(y); + if (x_bound->lower() >= 0 && x_bound->lower_instr() == NULL && y->as_ArrayLength() != NULL) { + _bound = new Bound(0, NULL, -1, y); + } else { + _bound = new Bound(); + } + } else if (!x->as_Constant() || !y->as_Constant()) { + assert(!x->as_Constant() || !y->as_Constant(), "One of the operands must be non-constant!"); + if (((x->as_Constant() || y->as_Constant()) && (ao->op() == Bytecodes::_iadd)) || (y->as_Constant() && ao->op() == Bytecodes::_isub)) { + assert(ao->op() == Bytecodes::_iadd || ao->op() == Bytecodes::_isub, "Operand must be iadd or isub"); + + if (y->as_Constant()) { + Value tmp = x; + x = y; + y = tmp; + } + assert(x->as_Constant()->type()->as_IntConstant(), "Constant must be int constant!"); + + // Constant now in x + int const_value = x->as_Constant()->type()->as_IntConstant()->value(); + if (ao->op() == Bytecodes::_iadd || const_value != min_jint) { + if (ao->op() == Bytecodes::_isub) { + const_value = -const_value; + } + + Bound * bound = _rce->get_bound(y); + if (bound->has_upper() && bound->has_lower()) { + int new_lower = bound->lower() + const_value; + jlong new_lowerl = ((jlong)bound->lower()) + const_value; + int new_upper = bound->upper() + const_value; + jlong new_upperl = ((jlong)bound->upper()) + const_value; + + if (((jlong)new_lower) == new_lowerl && ((jlong)new_upper == new_upperl)) { + Bound *newBound = new Bound(new_lower, bound->lower_instr(), new_upper, bound->upper_instr()); + _bound = newBound; + } else { + // overflow + _bound = new Bound(); + } + } else { + _bound = new Bound(); + } + } else { + _bound = new Bound(); + } + } else { + Bound *bound = _rce->get_bound(x); + if (ao->op() == Bytecodes::_isub) { + if (bound->lower_instr() == y) { + _bound = new Bound(Instruction::geq, NULL, bound->lower()); + } else { + _bound = new Bound(); + } + } else { + _bound = new Bound(); + } + } + } +} + +// IfOp +void RangeCheckEliminator::Visitor::do_IfOp(IfOp *ifOp) +{ + if (ifOp->tval()->type()->as_IntConstant() && ifOp->fval()->type()->as_IntConstant()) { + int min = ifOp->tval()->type()->as_IntConstant()->value(); + int max = ifOp->fval()->type()->as_IntConstant()->value(); + if (min > max) { + // min ^= max ^= min ^= max; + int tmp = min; + min = max; + max = tmp; + } + _bound = new Bound(min, NULL, max, NULL); + } +} + +// Get bound. Returns the current bound on Value v. Normally this is the topmost element on the bound stack. +RangeCheckEliminator::Bound *RangeCheckEliminator::get_bound(Value v) { + // Wrong type or NULL -> No bound + if (!v || (!v->type()->as_IntType() && !v->type()->as_ObjectType())) return NULL; + + if (!_bounds[v->id()]) { + // First (default) bound is calculated + // Create BoundStack + _bounds[v->id()] = new BoundStack(); + _visitor.clear_bound(); + Value visit_value = v; + visit_value->visit(&_visitor); + Bound *bound = _visitor.bound(); + if (bound) { + _bounds[v->id()]->push(bound); + } + if (_bounds[v->id()]->length() == 0) { + assert(!(v->as_Constant() && v->type()->as_IntConstant()), "constants not handled here"); + _bounds[v->id()]->push(new Bound()); + } + } else if (_bounds[v->id()]->length() == 0) { + // To avoid endless loops, bound is currently in calculation -> nothing known about it + return new Bound(); + } + + // Return bound + return _bounds[v->id()]->top(); +} + +// Update bound +void RangeCheckEliminator::update_bound(IntegerStack &pushed, Value v, Instruction::Condition cond, Value value, int constant) { + if (cond == Instruction::gtr) { + cond = Instruction::geq; + constant++; + } else if (cond == Instruction::lss) { + cond = Instruction::leq; + constant--; + } + Bound *bound = new Bound(cond, value, constant); + update_bound(pushed, v, bound); +} + +// Checks for loop invariance. Returns true if the instruction is outside of the loop which is identified by loop_header. +bool RangeCheckEliminator::loop_invariant(BlockBegin *loop_header, Instruction *instruction) { + assert(loop_header, "Loop header must not be null!"); + if (!instruction) return true; + return instruction->dominator_depth() < loop_header->dominator_depth(); +} + +// Update bound. Pushes a new bound onto the stack. Tries to do a conjunction with the current bound. +void RangeCheckEliminator::update_bound(IntegerStack &pushed, Value v, Bound *bound) { + if (v->as_Constant()) { + // No bound update for constants + return; + } + if (!_bounds[v->id()]) { + get_bound(v); + assert(_bounds[v->id()], "Now Stack must exist"); + } + Bound *top = NULL; + if (_bounds[v->id()]->length() > 0) { + top = _bounds[v->id()]->top(); + } + if (top) { + bound->and_op(top); + } + _bounds[v->id()]->push(bound); + pushed.append(v->id()); +} + +// Add instruction + integer for in block motion +void RangeCheckEliminator::add_access_indexed_info(InstructionList &indices, int integer, Value instruction, AccessIndexed *ai) { + int id = instruction->id(); + AccessIndexedInfo *aii = _access_indexed_info[id]; + if (aii == NULL) { + aii = new AccessIndexedInfo(); + _access_indexed_info[id] = aii; + indices.append(instruction); + aii->_min = integer; + aii->_max = integer; + aii->_list = new AccessIndexedList(); + } else if (integer >= aii->_min && integer <= aii->_max) { + remove_range_check(ai); + return; + } + aii->_min = MIN2(aii->_min, integer); + aii->_max = MAX2(aii->_max, integer); + aii->_list->append(ai); +} + +// In block motion. Tries to reorder checks in order to reduce some of them. +// Example: +// a[i] = 0; +// a[i+2] = 0; +// a[i+1] = 0; +// In this example the check for a[i+1] would be considered as unnecessary during the first iteration. +// After this i is only checked once for i >= 0 and i+2 < a.length before the first array access. If this +// check fails, deoptimization is called. +void RangeCheckEliminator::in_block_motion(BlockBegin *block, AccessIndexedList &accessIndexed, InstructionList &arrays) { + InstructionList indices; + Constant *c; + ArithmeticOp *ao; + + // Now iterate over all arrays + for (int i=0; iarray() != array || !ai->check_flag(Instruction::NeedsRangeCheckFlag)) continue; + + Value index = ai->index(); + if (c = index->as_Constant()) { + int constant_value = c->type()->as_IntConstant()->value(); + if (constant_value >= 0) { + if (constant_value <= max_constant) { + // No range check needed for this + remove_range_check(ai); + } else { + max_constant = constant_value; + list_constant.append(ai); + } + } + } else { + int last_integer = 0; + Instruction *last_instruction = index; + int base = 0; + + while ((ao = index->as_ArithmeticOp()) && (ao->x()->as_Constant() || ao->y()->as_Constant()) && (ao->op() == Bytecodes::_iadd || ao->op() == Bytecodes::_isub)) { + c = ao->y()->as_Constant(); + Instruction *other = ao->x(); + if (!c && ao->op() == Bytecodes::_iadd) { + c = ao->x()->as_Constant(); + other = ao->y(); + } + + if (c) { + int value = c->type()->as_IntConstant()->value(); + if (value != min_jint) { + if (ao->op() == Bytecodes::_isub) { + value = -value; + } + base += value; + last_integer = base; + last_instruction = other; + } + index = other; + } else { + break; + } + } + add_access_indexed_info(indices, last_integer, last_instruction, ai); + } + } + + // Iterate over all different indices + if (_optimistic) { + for (int i=0; iid()]; + assert(info != NULL, "Info must not be null"); + + // if idx < 0, max > 0, max + idx may fall between 0 and + // length-1 and if min < 0, min + idx may overflow and be >= + // 0. The predicate wouldn't trigger but some accesses could + // be with a negative index. This test guarantees that for the + // min and max value that are kept the predicate can't let + // some incorrect accesses happen. + bool range_cond = (info->_max < 0 || info->_max + min_jint <= info->_min); + + // Generate code only if more than 2 range checks can be eliminated because of that. + // 2 because at least 2 comparisons are done + if (info->_list->length() > 2 && range_cond) { + AccessIndexed *first = info->_list->at(0); + Instruction *insert_position = first->prev(); + assert(insert_position->next() == first, "prev was calculated"); + ValueStack *state = first->state_before(); + + // Load min Constant + Constant *min_constant = NULL; + if (info->_min != 0) { + min_constant = new Constant(new IntConstant(info->_min)); + NOT_PRODUCT(min_constant->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(min_constant); + } + + // Load max Constant + Constant *max_constant = NULL; + if (info->_max != 0) { + max_constant = new Constant(new IntConstant(info->_max)); + NOT_PRODUCT(max_constant->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(max_constant); + } + + // Load array length + Value length_instr = first->length(); + if (!length_instr) { + ArrayLength *length = new ArrayLength(array, first->state_before()->copy()); + length->set_exception_state(length->state_before()); + length->set_flag(Instruction::DeoptimizeOnException, true); + insert_position = insert_position->insert_after_same_bci(length); + length_instr = length; + } + + // Calculate lower bound + Instruction *lower_compare = index_instruction; + if (min_constant) { + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, min_constant, lower_compare, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + lower_compare = ao; + } + + // Calculate upper bound + Instruction *upper_compare = index_instruction; + if (max_constant) { + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, max_constant, upper_compare, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + upper_compare = ao; + } + + // Trick with unsigned compare is done + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(upper_compare, Instruction::aeq, true, length_instr, state->copy()); + NOT_PRODUCT(deoptimize->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(deoptimize); + Constant *constantMinus1 = new Constant(new IntConstant(-1)); + insert_position = insert_position->insert_after_same_bci(constantMinus1); + RangeCheckPredicate *deoptimize2 = new RangeCheckPredicate(lower_compare, Instruction::leq, true, constantMinus1, state->copy()); + insert_position = insert_position->insert_after_same_bci(deoptimize2); + for (int j = 0; j_list->length(); j++) { + AccessIndexed *ai = info->_list->at(j); + remove_range_check(ai); + } + } + _access_indexed_info[index_instruction->id()] = NULL; + } + indices.clear(); + + if (list_constant.length() > 1) { + AccessIndexed *first = list_constant.at(0); + Instruction *insert_position = first->prev(); + ValueStack *state = first->state_before(); + // Load max Constant + Constant *constant = new Constant(new IntConstant(max_constant)); + NOT_PRODUCT(constant->set_printable_bci(first->printable_bci())); + insert_position = insert_position->insert_after(constant); + Instruction *compare_instr = constant; + Value length_instr = first->length(); + if (!length_instr) { + ArrayLength *length = new ArrayLength(array, state->copy()); + length->set_exception_state(length->state_before()); + length->set_flag(Instruction::DeoptimizeOnException, true); + insert_position = insert_position->insert_after_same_bci(length); + length_instr = length; + } + // Compare for greater or equal to array length + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(compare_instr, Instruction::geq, true, length_instr, state->copy()); + insert_position = insert_position->insert_after_same_bci(deoptimize); + for (int j = 0; jas_AccessIndexed() != NULL); + cur = cur->next(); + } + + BlockList *dominates = block->dominates(); + for (int i=0; ilength(); i++) { + BlockBegin *next = dominates->at(i); + process |= set_process_block_flags(next); + } + + if (!process) { + block->set(BlockBegin::donot_eliminate_range_checks); + } + return process; +} + +bool RangeCheckEliminator::is_ok_for_deoptimization(Instruction *insert_position, Instruction *array_instr, Instruction *length_instr, Instruction *lower_instr, int lower, Instruction *upper_instr, int upper) { + bool upper_check = true; + assert(lower_instr || lower >= 0, "If no lower_instr present, lower must be greater 0"); + assert(!lower_instr || lower_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + assert(!upper_instr || upper_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + assert(array_instr, "Array instruction must exist"); + assert(array_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + assert(!length_instr || length_instr->dominator_depth() <= insert_position->dominator_depth(), "Dominator depth must be smaller"); + + if (upper_instr && upper_instr->as_ArrayLength() && upper_instr->as_ArrayLength()->array() == array_instr) { + // static check + if (upper >= 0) return false; // would always trigger a deopt: + // array_length + x >= array_length, x >= 0 is always true + upper_check = false; + } + if (lower_instr && lower_instr->as_ArrayLength() && lower_instr->as_ArrayLength()->array() == array_instr) { + if (lower > 0) return false; + } + // No upper check required -> skip + if (upper_check && upper_instr && upper_instr->type()->as_ObjectType() && upper_instr == array_instr) { + // upper_instr is object means that the upper bound is the length + // of the upper_instr. + return false; + } + return true; +} + +// Insert deoptimization, returns true if sucessful or false if range check should not be removed +void RangeCheckEliminator::insert_deoptimization(ValueStack *state, Instruction *insert_position, Instruction *array_instr, Instruction *length_instr, Instruction *lower_instr, int lower, Instruction *upper_instr, int upper, AccessIndexed *ai) { + assert(is_ok_for_deoptimization(insert_position, array_instr, length_instr, lower_instr, lower, upper_instr, upper), "should have been tested before"); + bool upper_check = !(upper_instr && upper_instr->as_ArrayLength() && upper_instr->as_ArrayLength()->array() == array_instr); + + if (lower_instr) { + assert(!lower_instr->type()->as_ObjectType(), "Must not be object type"); + if (lower == 0) { + // Constant0 + Constant *constant0 = new Constant(new IntConstant(0)); + NOT_PRODUCT(constant0->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(constant0); + // Compare for greater or equal 0 + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(lower_instr, Instruction::lss, true, constant0, state->copy()); + insert_position = insert_position->insert_after_same_bci(deoptimize); + } else if (lower > 0) { + // Constant0 + Constant *constant0 = new Constant(new IntConstant(0)); + NOT_PRODUCT(constant0->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(constant0); + Constant *constant = new Constant(new IntConstant(lower)); + insert_position = insert_position->insert_after_same_bci(constant); + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, constant, lower_instr, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + // Compare for smaller 0 + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(ao, Instruction::lss, true, constant0, state->copy()); + insert_position = insert_position->insert_after_same_bci(deoptimize); + } else { + assert(lower < 0, ""); + // Add 1 + lower++; + lower = -lower; + Constant *constant = new Constant(new IntConstant(lower)); + NOT_PRODUCT(constant->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(constant); + // Compare for smaller or equal 0 + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(lower_instr, Instruction::leq, true, constant, state->copy()); + insert_position = insert_position->insert_after_same_bci(deoptimize); + } + } + + // We need to know length of array + if (!length_instr) { + // Load length if necessary + ArrayLength *length = new ArrayLength(array_instr, state->copy()); + NOT_PRODUCT(length->set_printable_bci(ai->printable_bci())); + length->set_exception_state(length->state_before()); + length->set_flag(Instruction::DeoptimizeOnException, true); + insert_position = insert_position->insert_after(length); + length_instr = length; + } + + // No upper check required -> skip + if (!upper_check) return; + + if (!upper_instr) { + Constant *constant = new Constant(new IntConstant(upper)); + NOT_PRODUCT(constant->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(constant); + // Compare for geq array.length + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(constant, Instruction::geq, true, length_instr, state->copy()); + insert_position = insert_position->insert_after_same_bci(deoptimize); + } else { + if (upper_instr->type()->as_ObjectType()) { + assert(state, "must not be null"); + assert(upper_instr != array_instr, "should be"); + ArrayLength *length = new ArrayLength(upper_instr, state->copy()); + NOT_PRODUCT(length->set_printable_bci(ai->printable_bci())); + length->set_flag(Instruction::DeoptimizeOnException, true); + length->set_exception_state(length->state_before()); + insert_position = insert_position->insert_after(length); + upper_instr = length; + } + assert(upper_instr->type()->as_IntType(), "Must not be object type!"); + + if (upper == 0) { + // Compare for geq array.length + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(upper_instr, Instruction::geq, true, length_instr, state->copy()); + NOT_PRODUCT(deoptimize->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(deoptimize); + } else if (upper < 0) { + Constant *constant = new Constant(new IntConstant(upper)); + NOT_PRODUCT(constant->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(constant); + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, constant, upper_instr, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + // Compare for geq array.length + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(ao, Instruction::geq, true, length_instr, state->copy()); + NOT_PRODUCT(deoptimize->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after_same_bci(deoptimize); + } else { + assert(upper > 0, ""); + upper = -upper; + Constant *constant = new Constant(new IntConstant(upper)); + NOT_PRODUCT(constant->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after(constant); + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, constant, length_instr, false, NULL); + insert_position = insert_position->insert_after_same_bci(ao); + // Compare for geq array.length + RangeCheckPredicate *deoptimize = new RangeCheckPredicate(upper_instr, Instruction::geq, true, ao, state->copy()); + NOT_PRODUCT(deoptimize->set_printable_bci(ai->printable_bci())); + insert_position = insert_position->insert_after_same_bci(deoptimize); + } + } +} + +// Add if condition +void RangeCheckEliminator::add_if_condition(IntegerStack &pushed, Value x, Value y, Instruction::Condition condition) { + if (y->as_Constant()) return; + + int const_value = 0; + Value instr_value = x; + Constant *c; + ArithmeticOp *ao; + + if (c = x->as_Constant()) { + const_value = c->type()->as_IntConstant()->value(); + instr_value = NULL; + } else if ((ao = x->as_ArithmeticOp()) && (!ao->x()->as_Constant() || !ao->y()->as_Constant()) && ((ao->op() == Bytecodes::_isub && ao->y()->as_Constant()) || ao->op() == Bytecodes::_iadd)) { + assert(!ao->x()->as_Constant() || !ao->y()->as_Constant(), "At least one operator must be non-constant!"); + assert(ao->op() == Bytecodes::_isub || ao->op() == Bytecodes::_iadd, "Operation has to be add or sub!"); + if (c = ao->x()->as_Constant()) { + const_value = c->type()->as_IntConstant()->value(); + instr_value = ao->y(); + } else if (c = ao->y()->as_Constant()) { + const_value = c->type()->as_IntConstant()->value(); + instr_value = ao->x(); + } + if (ao->op() == Bytecodes::_isub) { + assert(ao->y()->as_Constant(), "1 - x not supported, only x - 1 is valid!"); + if (const_value > min_jint) { + const_value = -const_value; + } else { + const_value = 0; + instr_value = x; + } + } + } + + update_bound(pushed, y, condition, instr_value, const_value); +} + +// Process If +void RangeCheckEliminator::process_if(IntegerStack &pushed, BlockBegin *block, If *cond) { + // Only if we are direct true / false successor and NOT both ! (even this may occur) + if ((cond->tsux() == block || cond->fsux() == block) && cond->tsux() != cond->fsux()) { + Instruction::Condition condition = cond->cond(); + if (cond->fsux() == block) { + condition = Instruction::negate(condition); + } + Value x = cond->x(); + Value y = cond->y(); + if (x->type()->as_IntType() && y->type()->as_IntType()) { + add_if_condition(pushed, y, x, condition); + add_if_condition(pushed, x, y, Instruction::mirror(condition)); + } + } +} + +// Process access indexed +void RangeCheckEliminator::process_access_indexed(BlockBegin *loop_header, BlockBegin *block, AccessIndexed *ai) { + TRACE_RANGE_CHECK_ELIMINATION(tty->fill_to(block->dominator_depth()*2)); + TRACE_RANGE_CHECK_ELIMINATION(tty->print_cr("Access indexed: index=%d length=%d", ai->index()->id(), ai->length())); + + if (ai->check_flag(Instruction::NeedsRangeCheckFlag)) { + Bound *index_bound = get_bound(ai->index()); + if (!index_bound->has_lower() || !index_bound->has_upper()) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Index instruction %d has no lower and/or no upper bound!", ai->index()->id())); + return; + } + + Bound *array_bound; + if (ai->length()) { + array_bound = get_bound(ai->length()); + } else { + array_bound = get_bound(ai->array()); + } + + if (in_array_bound(index_bound, ai->array()) || + (index_bound && array_bound && index_bound->is_smaller(array_bound) && !index_bound->lower_instr() && index_bound->lower() >= 0)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Bounds check for instruction %d in block B%d can be fully eliminated!", ai->id(), ai->block()->block_id())); + + remove_range_check(ai); + } else if (_optimistic && loop_header) { + assert(ai->array(), "Array must not be null!"); + assert(ai->index(), "Index must not be null!"); + + // Array instruction + Instruction *array_instr = ai->array(); + if (!loop_invariant(loop_header, array_instr)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Array %d is not loop invariant to header B%d", ai->array()->id(), loop_header->block_id())); + return; + } + + // Lower instruction + Value index_instr = ai->index(); + Value lower_instr = index_bound->lower_instr(); + if (!loop_invariant(loop_header, lower_instr)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Lower instruction %d not loop invariant!", lower_instr->id())); + return; + } + if (!lower_instr && index_bound->lower() < 0) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Lower bound smaller than 0 (%d)!", index_bound->lower())); + return; + } + + // Upper instruction + Value upper_instr = index_bound->upper_instr(); + if (!loop_invariant(loop_header, upper_instr)) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Upper instruction %d not loop invariant!", upper_instr->id())); + return; + } + + // Length instruction + Value length_instr = ai->length(); + if (!loop_invariant(loop_header, length_instr)) { + // Generate length instruction yourself! + length_instr = NULL; + } + + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("LOOP INVARIANT access indexed %d found in block B%d!", ai->id(), ai->block()->block_id())); + + BlockBegin *pred_block = loop_header->dominator(); + assert(pred_block != NULL, "Every loop header has a dominator!"); + BlockEnd *pred_block_end = pred_block->end(); + Instruction *insert_position = pred_block_end->prev(); + ValueStack *state = pred_block_end->state_before(); + if (pred_block_end->as_Goto() && state == NULL) state = pred_block_end->state(); + assert(state, "State must not be null"); + + // Add deoptimization to dominator of loop header + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Inserting deopt at bci %d in block B%d!", state->bci(), insert_position->block()->block_id())); + + if (!is_ok_for_deoptimization(insert_position, array_instr, length_instr, lower_instr, index_bound->lower(), upper_instr, index_bound->upper())) { + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Could not eliminate because of static analysis!")); + return; + } + + insert_deoptimization(state, insert_position, array_instr, length_instr, lower_instr, index_bound->lower(), upper_instr, index_bound->upper(), ai); + + // Finally remove the range check! + remove_range_check(ai); + } + } +} + +void RangeCheckEliminator::remove_range_check(AccessIndexed *ai) { + ai->set_flag(Instruction::NeedsRangeCheckFlag, false); + // no range check, no need for the length instruction anymore + ai->clear_length(); + + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(ai->dominator_depth()*2); + tty->print_cr("Range check for instruction %d eliminated!", ai->id()); + ); + + ASSERT_RANGE_CHECK_ELIMINATION( + Value array_length = ai->length(); + if (!array_length) { + array_length = ai->array(); + assert(array_length->type()->as_ObjectType(), "Has to be object type!"); + } + int cur_constant = -1; + Value cur_value = array_length; + if (cur_value->type()->as_IntConstant()) { + cur_constant += cur_value->type()->as_IntConstant()->value(); + cur_value = NULL; + } + Bound *new_index_bound = new Bound(0, NULL, cur_constant, cur_value); + add_assertions(new_index_bound, ai->index(), ai); + ); +} + +// Calculate bounds for instruction in this block and children blocks in the dominator tree +void RangeCheckEliminator::calc_bounds(BlockBegin *block, BlockBegin *loop_header) { + // Ensures a valid loop_header + assert(!loop_header || loop_header->is_set(BlockBegin::linear_scan_loop_header_flag), "Loop header has to be real !"); + + // Tracing output + TRACE_RANGE_CHECK_ELIMINATION( + tty->fill_to(block->dominator_depth()*2); + tty->print_cr("Block B%d", block->block_id()); + ); + + // Pushed stack for conditions + IntegerStack pushed; + // Process If + BlockBegin *parent = block->dominator(); + If *cond; + if (parent && (cond = parent->end()->as_If())) { + process_if(pushed, block, cond); + } + + // Interate over current block + InstructionList arrays; + AccessIndexedList accessIndexed; + AccessIndexed *ai; + Instruction *cur = block; + + while (cur) { + // Ensure cur wasn't inserted during the elimination + if (cur->id() < this->_bounds.length()) { + // Process only if it is an access indexed instruction + if (ai = cur->as_AccessIndexed()) { + process_access_indexed(loop_header, block, ai); + accessIndexed.append(ai); + if (!arrays.contains(ai->array())) { + arrays.append(ai->array()); + } + Bound *b = get_bound(ai->index()); + if (!b->lower_instr()) { + // Lower bound is constant + update_bound(pushed, ai->index(), Instruction::geq, NULL, 0); + } + if (!b->has_upper()) { + if (ai->length() && ai->length()->type()->as_IntConstant()) { + int value = ai->length()->type()->as_IntConstant()->value(); + update_bound(pushed, ai->index(), Instruction::lss, NULL, value); + } else { + // Has no upper bound + Instruction *instr = ai->length(); + if (!instr) instr = ai->array(); + update_bound(pushed, ai->index(), Instruction::lss, instr, 0); + } + } + } + } + cur = cur->next(); + } + + // Output current condition stack + TRACE_RANGE_CHECK_ELIMINATION(dump_condition_stack(block)); + + // Do in block motion of range checks + in_block_motion(block, accessIndexed, arrays); + + // Call all dominated blocks + for (int i=0; idominates()->length(); i++) { + BlockBegin *next = block->dominates()->at(i); + if (!next->is_set(BlockBegin::donot_eliminate_range_checks)) { + // if current block is a loop header and: + // - next block belongs to the same loop + // or + // - next block belongs to an inner loop + // then current block is the loop header for next block + if (block->is_set(BlockBegin::linear_scan_loop_header_flag) && (block->loop_index() == next->loop_index() || next->loop_depth() > block->loop_depth())) { + calc_bounds(next, block); + } else { + calc_bounds(next, loop_header); + } + } + } + + // Reset stack + for (int i=0; ipop(); + } +} + +#ifndef PRODUCT + +// Dump condition stack +void RangeCheckEliminator::dump_condition_stack(BlockBegin *block) { + for (int i=0; i<_ir->linear_scan_order()->length(); i++) { + BlockBegin *cur_block = _ir->linear_scan_order()->at(i); + Instruction *instr = cur_block; + for_each_phi_fun(cur_block, phi, + BoundStack *bound_stack = _bounds.at(phi->id()); + if (bound_stack && bound_stack->length() > 0) { + Bound *bound = bound_stack->top(); + if ((bound->has_lower() || bound->has_upper()) && (bound->lower_instr() != phi || bound->upper_instr() != phi || bound->lower() != 0 || bound->upper() != 0)) { + TRACE_RANGE_CHECK_ELIMINATION(tty->fill_to(2*block->dominator_depth()); + tty->print("i%d", phi->id()); + tty->print(": "); + bound->print(); + tty->print_cr("");); + } + }); + + while (!instr->as_BlockEnd()) { + if (instr->id() < _bounds.length()) { + BoundStack *bound_stack = _bounds.at(instr->id()); + if (bound_stack && bound_stack->length() > 0) { + Bound *bound = bound_stack->top(); + if ((bound->has_lower() || bound->has_upper()) && (bound->lower_instr() != instr || bound->upper_instr() != instr || bound->lower() != 0 || bound->upper() != 0)) { + TRACE_RANGE_CHECK_ELIMINATION(tty->fill_to(2*block->dominator_depth()); + tty->print("i%d", instr->id()); + tty->print(": "); + bound->print(); + tty->print_cr("");); + } + } + } + instr = instr->next(); + } + } +} + +#endif + +// Verification or the IR +RangeCheckEliminator::Verification::Verification(IR *ir) : _used(BlockBegin::number_of_blocks(), false) { + this->_ir = ir; + ir->iterate_linear_scan_order(this); +} + +// Verify this block +void RangeCheckEliminator::Verification::block_do(BlockBegin *block) { + If *cond = block->end()->as_If(); + // Watch out: tsux and fsux can be the same! + if (block->number_of_sux() > 1) { + for (int i=0; inumber_of_sux(); i++) { + BlockBegin *sux = block->sux_at(i); + BlockBegin *pred = NULL; + for (int j=0; jnumber_of_preds(); j++) { + BlockBegin *cur = sux->pred_at(j); + assert(cur != NULL, "Predecessor must not be null"); + if (!pred) { + pred = cur; + } + assert(cur == pred, "Block must not have more than one predecessor if its predecessor has more than one successor"); + } + assert(sux->number_of_preds() >= 1, "Block must have at least one predecessor"); + assert(sux->pred_at(0) == block, "Wrong successor"); + } + } + + BlockBegin *dominator = block->dominator(); + if (dominator) { + assert(block != _ir->start(), "Start block must not have a dominator!"); + assert(can_reach(dominator, block), "Dominator can't reach his block !"); + assert(can_reach(_ir->start(), dominator), "Dominator is unreachable !"); + assert(!can_reach(_ir->start(), block, dominator), "Wrong dominator ! Block can be reached anyway !"); + BlockList *all_blocks = _ir->linear_scan_order(); + for (int i=0; ilength(); i++) { + BlockBegin *cur = all_blocks->at(i); + if (cur != dominator && cur != block) { + assert(can_reach(dominator, block, cur), "There has to be another dominator!"); + } + } + } else { + assert(block == _ir->start(), "Only start block must not have a dominator"); + } + + if (block->is_set(BlockBegin::linear_scan_loop_header_flag)) { + int loop_index = block->loop_index(); + BlockList *all_blocks = _ir->linear_scan_order(); + assert(block->number_of_preds() >= 1, "Block must have at least one predecessor"); + assert(!block->is_set(BlockBegin::exception_entry_flag), "Loop header must not be exception handler!"); + // Sometimes, the backbranch comes from an exception handler. In + // this case, loop indexes/loop depths may not appear correct. + bool loop_through_xhandler = false; + for (int i = 0; i < block->number_of_exception_handlers(); i++) { + BlockBegin *xhandler = block->exception_handler_at(i); + for (int j = 0; j < block->number_of_preds(); j++) { + if (dominates(xhandler, block->pred_at(j)) || xhandler == block->pred_at(j)) { + loop_through_xhandler = true; + } + } + } + + for (int i=0; inumber_of_sux(); i++) { + BlockBegin *sux = block->sux_at(i); + assert(sux->loop_depth() != block->loop_depth() || sux->loop_index() == block->loop_index() || loop_through_xhandler, "Loop index has to be same"); + assert(sux->loop_depth() == block->loop_depth() || sux->loop_index() != block->loop_index(), "Loop index has to be different"); + } + + for (int i=0; ilength(); i++) { + BlockBegin *cur = all_blocks->at(i); + if (cur->loop_index() == loop_index && cur != block) { + assert(dominates(block->dominator(), cur), "Dominator of loop header must dominate all loop blocks"); + } + } + } + + Instruction *cur = block; + while (cur) { + assert(cur->block() == block, "Block begin has to be set correctly!"); + cur = cur->next(); + } +} + +// Loop header must dominate all loop blocks +bool RangeCheckEliminator::Verification::dominates(BlockBegin *dominator, BlockBegin *block) { + BlockBegin *cur = block->dominator(); + while (cur && cur != dominator) { + cur = cur->dominator(); + } + return cur == dominator; +} + +// Try to reach Block end beginning in Block start and not using Block dont_use +bool RangeCheckEliminator::Verification::can_reach(BlockBegin *start, BlockBegin *end, BlockBegin *dont_use /* = NULL */) { + if (start == end) return start != dont_use; + // Simple BSF from start to end + // BlockBeginList _current; + for (int i=0; i<_used.length(); i++) { + _used[i] = false; + } + _current.truncate(0); + _successors.truncate(0); + if (start != dont_use) { + _current.push(start); + _used[start->block_id()] = true; + } + + // BlockBeginList _successors; + while (_current.length() > 0) { + BlockBegin *cur = _current.pop(); + // Add exception handlers to list + for (int i=0; inumber_of_exception_handlers(); i++) { + BlockBegin *xhandler = cur->exception_handler_at(i); + _successors.push(xhandler); + // Add exception handlers of _successors to list + for (int j=0; jnumber_of_exception_handlers(); j++) { + BlockBegin *sux_xhandler = xhandler->exception_handler_at(j); + _successors.push(sux_xhandler); + } + } + // Add normal _successors to list + for (int i=0; inumber_of_sux(); i++) { + BlockBegin *sux = cur->sux_at(i); + _successors.push(sux); + // Add exception handlers of _successors to list + for (int j=0; jnumber_of_exception_handlers(); j++) { + BlockBegin *xhandler = sux->exception_handler_at(j); + _successors.push(xhandler); + } + } + for (int i=0; i<_successors.length(); i++) { + BlockBegin *sux = _successors[i]; + assert(sux != NULL, "Successor must not be NULL!"); + if (sux == end) { + return true; + } + if (sux != dont_use && !_used[sux->block_id()]) { + _used[sux->block_id()] = true; + _current.push(sux); + } + } + _successors.truncate(0); + } + + return false; +} + +// Bound +RangeCheckEliminator::Bound::~Bound() { +} + +// Bound constructor +RangeCheckEliminator::Bound::Bound() { + init(); + this->_lower = min_jint; + this->_upper = max_jint; + this->_lower_instr = NULL; + this->_upper_instr = NULL; +} + +// Bound constructor +RangeCheckEliminator::Bound::Bound(int lower, Value lower_instr, int upper, Value upper_instr) { + init(); + assert(!lower_instr || !lower_instr->as_Constant() || !lower_instr->type()->as_IntConstant(), "Must not be constant!"); + assert(!upper_instr || !upper_instr->as_Constant() || !upper_instr->type()->as_IntConstant(), "Must not be constant!"); + this->_lower = lower; + this->_upper = upper; + this->_lower_instr = lower_instr; + this->_upper_instr = upper_instr; +} + +// Bound constructor +RangeCheckEliminator::Bound::Bound(Instruction::Condition cond, Value v, int constant) { + assert(!v || (v->type() && (v->type()->as_IntType() || v->type()->as_ObjectType())), "Type must be array or integer!"); + assert(!v || !v->as_Constant() || !v->type()->as_IntConstant(), "Must not be constant!"); + + init(); + if (cond == Instruction::eql) { + _lower = constant; + _lower_instr = v; + _upper = constant; + _upper_instr = v; + } else if (cond == Instruction::neq) { + _lower = min_jint; + _upper = max_jint; + _lower_instr = NULL; + _upper_instr = NULL; + if (v == NULL) { + if (constant == min_jint) { + _lower++; + } + if (constant == max_jint) { + _upper--; + } + } + } else if (cond == Instruction::geq) { + _lower = constant; + _lower_instr = v; + _upper = max_jint; + _upper_instr = NULL; + } else if (cond == Instruction::leq) { + _lower = min_jint; + _lower_instr = NULL; + _upper = constant; + _upper_instr = v; + } else { + ShouldNotReachHere(); + } +} + +// Set lower +void RangeCheckEliminator::Bound::set_lower(int value, Value v) { + assert(!v || !v->as_Constant() || !v->type()->as_IntConstant(), "Must not be constant!"); + this->_lower = value; + this->_lower_instr = v; +} + +// Set upper +void RangeCheckEliminator::Bound::set_upper(int value, Value v) { + assert(!v || !v->as_Constant() || !v->type()->as_IntConstant(), "Must not be constant!"); + this->_upper = value; + this->_upper_instr = v; +} + +// Add constant -> no overflow may occur +void RangeCheckEliminator::Bound::add_constant(int value) { + this->_lower += value; + this->_upper += value; +} + +// Init +void RangeCheckEliminator::Bound::init() { +} + +// or +void RangeCheckEliminator::Bound::or_op(Bound *b) { + // Watch out, bound is not guaranteed not to overflow! + // Update lower bound + if (_lower_instr != b->_lower_instr || (_lower_instr && _lower != b->_lower)) { + _lower_instr = NULL; + _lower = min_jint; + } else { + _lower = MIN2(_lower, b->_lower); + } + // Update upper bound + if (_upper_instr != b->_upper_instr || (_upper_instr && _upper != b->_upper)) { + _upper_instr = NULL; + _upper = max_jint; + } else { + _upper = MAX2(_upper, b->_upper); + } +} + +// and +void RangeCheckEliminator::Bound::and_op(Bound *b) { + // Update lower bound + if (_lower_instr == b->_lower_instr) { + _lower = MAX2(_lower, b->_lower); + } + if (b->has_lower()) { + bool set = true; + if (_lower_instr != NULL && b->_lower_instr != NULL) { + set = (_lower_instr->dominator_depth() > b->_lower_instr->dominator_depth()); + } + if (set) { + _lower = b->_lower; + _lower_instr = b->_lower_instr; + } + } + // Update upper bound + if (_upper_instr == b->_upper_instr) { + _upper = MIN2(_upper, b->_upper); + } + if (b->has_upper()) { + bool set = true; + if (_upper_instr != NULL && b->_upper_instr != NULL) { + set = (_upper_instr->dominator_depth() > b->_upper_instr->dominator_depth()); + } + if (set) { + _upper = b->_upper; + _upper_instr = b->_upper_instr; + } + } +} + +// has_upper +bool RangeCheckEliminator::Bound::has_upper() { + return _upper_instr != NULL || _upper < max_jint; +} + +// is_smaller +bool RangeCheckEliminator::Bound::is_smaller(Bound *b) { + if (b->_lower_instr != _upper_instr) { + return false; + } + return _upper < b->_lower; +} + +// has_lower +bool RangeCheckEliminator::Bound::has_lower() { + return _lower_instr != NULL || _lower > min_jint; +} + +// in_array_bound +bool RangeCheckEliminator::in_array_bound(Bound *bound, Value array){ + if (!bound) return false; + assert(array != NULL, "Must not be null!"); + assert(bound != NULL, "Must not be null!"); + ArrayLength *len; + if (bound->lower() >=0 && bound->lower_instr() == NULL && bound->upper() < 0 && bound->upper_instr() != NULL) { + if (bound->upper_instr() == array || ((len = bound->upper_instr()->as_ArrayLength()) && len->array() == array)) { + return true; + } + } + return false; +} + +// remove_lower +void RangeCheckEliminator::Bound::remove_lower() { + _lower = min_jint; + _lower_instr = NULL; +} + +// remove_upper +void RangeCheckEliminator::Bound::remove_upper() { + _upper = max_jint; + _upper_instr = NULL; +} + +// upper +int RangeCheckEliminator::Bound::upper() { + return _upper; +} + +// lower +int RangeCheckEliminator::Bound::lower() { + return _lower; +} + +// upper_instr +Value RangeCheckEliminator::Bound::upper_instr() { + return _upper_instr; +} + +// lower_instr +Value RangeCheckEliminator::Bound::lower_instr() { + return _lower_instr; +} + +// print +void RangeCheckEliminator::Bound::print() { + tty->print(""); + if (this->_lower_instr || this->_lower != min_jint) { + if (this->_lower_instr) { + tty->print("i%d", this->_lower_instr->id()); + if (this->_lower > 0) { + tty->print("+%d", _lower); + } + if (this->_lower < 0) { + tty->print("%d", _lower); + } + } else { + tty->print("%d", _lower); + } + tty->print(" <= "); + } + tty->print("x"); + if (this->_upper_instr || this->_upper != max_jint) { + tty->print(" <= "); + if (this->_upper_instr) { + tty->print("i%d", this->_upper_instr->id()); + if (this->_upper > 0) { + tty->print("+%d", _upper); + } + if (this->_upper < 0) { + tty->print("%d", _upper); + } + } else { + tty->print("%d", _upper); + } + } +} + +// Copy +RangeCheckEliminator::Bound *RangeCheckEliminator::Bound::copy() { + Bound *b = new Bound(); + b->_lower = _lower; + b->_lower_instr = _lower_instr; + b->_upper = _upper; + b->_upper_instr = _upper_instr; + return b; +} + +#ifndef PRODUCT +// Add assertion +void RangeCheckEliminator::Bound::add_assertion(Instruction *instruction, Instruction *position, int integer, Value instr, Instruction::Condition cond) { + Instruction *result = position; + Instruction *compare_with = NULL; + ValueStack *state = position->state_before(); + if (position->as_BlockEnd() && !position->as_Goto()) { + state = position->as_BlockEnd()->state_before(); + } + Instruction *instruction_before = position->prev(); + if (position->as_Return() && Compilation::current()->method()->is_synchronized() && instruction_before->as_MonitorExit()) { + instruction_before = instruction_before->prev(); + } + result = instruction_before; + // Load constant only if needed + Constant *constant = NULL; + if (integer != 0 || !instr) { + constant = new Constant(new IntConstant(integer)); + NOT_PRODUCT(constant->set_printable_bci(position->printable_bci())); + result = result->insert_after(constant); + compare_with = constant; + } + + if (instr) { + assert(instr->type()->as_ObjectType() || instr->type()->as_IntType(), "Type must be array or integer!"); + compare_with = instr; + // Load array length if necessary + Instruction *op = instr; + if (instr->type()->as_ObjectType()) { + assert(state, "must not be null"); + ArrayLength *length = new ArrayLength(instr, state->copy()); + NOT_PRODUCT(length->set_printable_bci(position->printable_bci())); + length->set_exception_state(length->state_before()); + result = result->insert_after(length); + op = length; + compare_with = length; + } + // Add operation only if necessary + if (constant) { + ArithmeticOp *ao = new ArithmeticOp(Bytecodes::_iadd, constant, op, false, NULL); + NOT_PRODUCT(ao->set_printable_bci(position->printable_bci())); + result = result->insert_after(ao); + compare_with = ao; + // TODO: Check that add operation does not overflow! + } + } + assert(compare_with != NULL, "You have to compare with something!"); + assert(instruction != NULL, "Instruction must not be null!"); + + if (instruction->type()->as_ObjectType()) { + // Load array length if necessary + Instruction *op = instruction; + assert(state, "must not be null"); + ArrayLength *length = new ArrayLength(instruction, state->copy()); + length->set_exception_state(length->state_before()); + NOT_PRODUCT(length->set_printable_bci(position->printable_bci())); + result = result->insert_after(length); + instruction = length; + } + + Assert *assert = new Assert(instruction, cond, false, compare_with); + NOT_PRODUCT(assert->set_printable_bci(position->printable_bci())); + result->insert_after(assert); +} + +// Add assertions +void RangeCheckEliminator::add_assertions(Bound *bound, Instruction *instruction, Instruction *position) { + // Add lower bound assertion + if (bound->has_lower()) { + bound->add_assertion(instruction, position, bound->lower(), bound->lower_instr(), Instruction::geq); + } + // Add upper bound assertion + if (bound->has_upper()) { + bound->add_assertion(instruction, position, bound->upper(), bound->upper_instr(), Instruction::leq); + } +} +#endif + --- /dev/null 2012-11-23 07:18:02.922988661 +0100 +++ new/src/share/vm/c1/c1_RangeCheckElimination.hpp 2013-02-20 17:41:14.290667702 +0100 @@ -0,0 +1,233 @@ +/* + * Copyright (c) 2012, Oracle and/or its affiliates. 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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA + * or visit www.oracle.com if you need additional information or have any + * questions. + * + */ + +#ifndef SHARE_VM_C1_C1_RANGECHECKELIMINATION_HPP +#define SHARE_VM_C1_C1_RANGECHECKELIMINATION_HPP + +#include "c1/c1_Instruction.hpp" + +// Base class for range check elimination +class RangeCheckElimination : AllStatic { +public: + static void eliminate(IR *ir); +}; + +// Implementation +class RangeCheckEliminator VALUE_OBJ_CLASS_SPEC { +private: + int _number_of_instructions; + bool _optimistic; // Insert predicates and deoptimize when they fail + IR *_ir; + + define_array(BlockBeginArray, BlockBegin*) + define_stack(BlockBeginList, BlockBeginArray) + define_stack(IntegerStack, intArray) + define_array(IntegerMap, IntegerStack*) + + class Verification : public _ValueObj /*VALUE_OBJ_CLASS_SPEC*/, public BlockClosure { + private: + IR *_ir; + boolArray _used; + BlockBeginList _current; + BlockBeginList _successors; + + public: + Verification(IR *ir); + virtual void block_do(BlockBegin *block); + bool can_reach(BlockBegin *start, BlockBegin *end, BlockBegin *dont_use = NULL); + bool dominates(BlockBegin *dominator, BlockBegin *block); + }; + +public: + // Bounds for an instruction in the form x + c which c integer + // constant and x another instruction + class Bound : public CompilationResourceObj { + private: + int _upper; + Value _upper_instr; + int _lower; + Value _lower_instr; + + public: + Bound(); + Bound(Value v); + Bound(Instruction::Condition cond, Value v, int constant = 0); + Bound(int lower, Value lower_instr, int upper, Value upper_instr); + ~Bound(); + +#ifndef PRODUCT + void add_assertion(Instruction *instruction, Instruction *position, int integer, Value instr, Instruction::Condition cond); +#endif + int upper(); + Value upper_instr(); + int lower(); + Value lower_instr(); + void print(); + bool check_no_overflow(int const_value); + void or_op(Bound *b); + void and_op(Bound *b); + bool has_upper(); + bool has_lower(); + void set_upper(int upper, Value upper_instr); + void set_lower(int lower, Value lower_instr); + bool is_smaller(Bound *b); + void remove_upper(); + void remove_lower(); + void add_constant(int value); + Bound *copy(); + + private: + void init(); + }; + + + class Visitor : public InstructionVisitor { + private: + Bound *_bound; + RangeCheckEliminator *_rce; + + public: + void set_range_check_eliminator(RangeCheckEliminator *rce) { _rce = rce; } + Bound *bound() const { return _bound; } + void clear_bound() { _bound = NULL; } + + protected: + // visitor functions + void do_Constant (Constant* x); + void do_IfOp (IfOp* x); + void do_LogicOp (LogicOp* x); + void do_ArithmeticOp (ArithmeticOp* x); + void do_Phi (Phi* x); + + void do_StoreField (StoreField* x) { /* nothing to do */ }; + void do_StoreIndexed (StoreIndexed* x) { /* nothing to do */ }; + void do_MonitorEnter (MonitorEnter* x) { /* nothing to do */ }; + void do_MonitorExit (MonitorExit* x) { /* nothing to do */ }; + void do_Invoke (Invoke* x) { /* nothing to do */ }; + void do_UnsafePutRaw (UnsafePutRaw* x) { /* nothing to do */ }; + void do_UnsafePutObject(UnsafePutObject* x) { /* nothing to do */ }; + void do_Intrinsic (Intrinsic* x) { /* nothing to do */ }; + void do_Local (Local* x) { /* nothing to do */ }; + void do_LoadField (LoadField* x) { /* nothing to do */ }; + void do_ArrayLength (ArrayLength* x) { /* nothing to do */ }; + void do_LoadIndexed (LoadIndexed* x) { /* nothing to do */ }; + void do_NegateOp (NegateOp* x) { /* nothing to do */ }; + void do_ShiftOp (ShiftOp* x) { /* nothing to do */ }; + void do_CompareOp (CompareOp* x) { /* nothing to do */ }; + void do_Convert (Convert* x) { /* nothing to do */ }; + void do_NullCheck (NullCheck* x) { /* nothing to do */ }; + void do_TypeCast (TypeCast* x) { /* nothing to do */ }; + void do_NewInstance (NewInstance* x) { /* nothing to do */ }; + void do_NewTypeArray (NewTypeArray* x) { /* nothing to do */ }; + void do_NewObjectArray (NewObjectArray* x) { /* nothing to do */ }; + void do_NewMultiArray (NewMultiArray* x) { /* nothing to do */ }; + void do_CheckCast (CheckCast* x) { /* nothing to do */ }; + void do_InstanceOf (InstanceOf* x) { /* nothing to do */ }; + void do_BlockBegin (BlockBegin* x) { /* nothing to do */ }; + void do_Goto (Goto* x) { /* nothing to do */ }; + void do_If (If* x) { /* nothing to do */ }; + void do_IfInstanceOf (IfInstanceOf* x) { /* nothing to do */ }; + void do_TableSwitch (TableSwitch* x) { /* nothing to do */ }; + void do_LookupSwitch (LookupSwitch* x) { /* nothing to do */ }; + void do_Return (Return* x) { /* nothing to do */ }; + void do_Throw (Throw* x) { /* nothing to do */ }; + void do_Base (Base* x) { /* nothing to do */ }; + void do_OsrEntry (OsrEntry* x) { /* nothing to do */ }; + void do_ExceptionObject(ExceptionObject* x) { /* nothing to do */ }; + void do_RoundFP (RoundFP* x) { /* nothing to do */ }; + void do_UnsafeGetRaw (UnsafeGetRaw* x) { /* nothing to do */ }; + void do_UnsafeGetObject(UnsafeGetObject* x) { /* nothing to do */ }; + void do_UnsafeGetAndSetObject(UnsafeGetAndSetObject* x) { /* nothing to do */ }; + void do_UnsafePrefetchRead (UnsafePrefetchRead* x) { /* nothing to do */ }; + void do_UnsafePrefetchWrite(UnsafePrefetchWrite* x) { /* nothing to do */ }; + void do_ProfileCall (ProfileCall* x) { /* nothing to do */ }; + void do_ProfileInvoke (ProfileInvoke* x) { /* nothing to do */ }; + void do_RuntimeCall (RuntimeCall* x) { /* nothing to do */ }; + void do_MemBar (MemBar* x) { /* nothing to do */ }; + void do_RangeCheckPredicate(RangeCheckPredicate* x) { /* nothing to do */ }; + void do_Assert (Assert* x) { /* nothing to do */ }; + }; + +#ifndef PRODUCT + void add_assertions(Bound *bound, Instruction *instruction, Instruction *position); +#endif + + define_array(BoundArray, Bound *) + define_stack(BoundStack, BoundArray) + define_array(BoundMap, BoundStack *) + define_array(AccessIndexedArray, AccessIndexed *) + define_stack(AccessIndexedList, AccessIndexedArray) + define_array(InstructionArray, Instruction *) + define_stack(InstructionList, InstructionArray) + + class AccessIndexedInfo : public CompilationResourceObj { + public: + AccessIndexedList *_list; + int _min; + int _max; + }; + + define_array(AccessIndexedInfoArray, AccessIndexedInfo *) + BoundMap _bounds; // Mapping from Instruction's id to current bound + AccessIndexedInfoArray _access_indexed_info; // Mapping from Instruction's id to AccessIndexedInfo for in block motion + Visitor _visitor; + +public: + RangeCheckEliminator(IR *ir); + + IR *ir() const { return _ir; } + + // Pass over the dominator tree to identify blocks where there's an oppportunity for optimization + bool set_process_block_flags(BlockBegin *block); + // The core of the optimization work: pass over the dominator tree + // to propagate bound information, insert predicate out of loops, + // eliminate bound checks when possible and perform in block motion + void calc_bounds(BlockBegin *block, BlockBegin *loop_header); + // reorder bound checks within a block in order to eliminate some of them + void in_block_motion(BlockBegin *block, AccessIndexedList &accessIndexed, InstructionList &arrays); + + // update/access current bound + void update_bound(IntegerStack &pushed, Value v, Instruction::Condition cond, Value value, int constant); + void update_bound(IntegerStack &pushed, Value v, Bound *bound); + Bound *get_bound(Value v); + + bool loop_invariant(BlockBegin *loop_header, Instruction *instruction); // check for loop invariance + void add_access_indexed_info(InstructionList &indices, int integer, Value instruction, AccessIndexed *ai); // record indexed access for in block motion + void remove_range_check(AccessIndexed *ai); // Mark this instructions as not needing a range check + void add_if_condition(IntegerStack &pushed, Value x, Value y, Instruction::Condition condition); // Update bound for an If + bool in_array_bound(Bound *bound, Value array); // Check whether bound is known to fall within array + void insert_deoptimization(ValueStack *state, Instruction *insert_position, Instruction *array_instr, // Add predicate + Instruction *length_instruction, Instruction *lower_instr, int lower, + Instruction *upper_instr, int upper, AccessIndexed *ai); + bool is_ok_for_deoptimization(Instruction *insert_position, Instruction *array_instr, // Can we safely add a predicate? + Instruction *length_instr, Instruction *lower_instr, + int lower, Instruction *upper_instr, int upper); + void process_if(IntegerStack &pushed, BlockBegin *block, If *cond); // process If Instruction + void process_access_indexed(BlockBegin *loop_header, BlockBegin *block, AccessIndexed *ai); // process indexed access + + void dump_condition_stack(BlockBegin *cur_block); + static void print_statistics(); +}; + +#endif // SHARE_VM_C1_C1_RANGECHECKELIMINATION_HPP