Home | History | Annotate | Download | only in optimizing 
      1 /*
      2  * Copyright (C) 2015 The Android Open Source Project
      3  *
      4  * Licensed under the Apache License, Version 2.0 (the "License");
      5  * you may not use this file except in compliance with the License.
      6  * You may obtain a copy of the License at
      7  *
      8  *      http://www.apache.org/licenses/LICENSE-2.0
      9  *
     10  * Unless required by applicable law or agreed to in writing, software
     11  * distributed under the License is distributed on an "AS IS" BASIS,
     12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     13  * See the License for the specific language governing permissions and
     14  * limitations under the License.
     15  */
     16 
     17 #include "pc_relative_fixups_x86.h"
     18 #include "code_generator_x86.h"
     19 #include "intrinsics_x86.h"
     20 
     21 namespace art {
     22 namespace x86 {
     23 
     24 /**
     25  * Finds instructions that need the constant area base as an input.
     26  */
     27 class PCRelativeHandlerVisitor : public HGraphVisitor {
     28  public:
     29   PCRelativeHandlerVisitor(HGraph* graph, CodeGenerator* codegen)
     30       : HGraphVisitor(graph),
     31         codegen_(down_cast<CodeGeneratorX86*>(codegen)),
     32         base_(nullptr) {}
     33 
     34   void MoveBaseIfNeeded() {
     35     if (base_ != nullptr) {
     36       // Bring the base closer to the first use (previously, it was in the
     37       // entry block) and relieve some pressure on the register allocator
     38       // while avoiding recalculation of the base in a loop.
     39       base_->MoveBeforeFirstUserAndOutOfLoops();
     40     }
     41   }
     42 
     43  private:
     44   void VisitAdd(HAdd* add) OVERRIDE {
     45     BinaryFP(add);
     46   }
     47 
     48   void VisitSub(HSub* sub) OVERRIDE {
     49     BinaryFP(sub);
     50   }
     51 
     52   void VisitMul(HMul* mul) OVERRIDE {
     53     BinaryFP(mul);
     54   }
     55 
     56   void VisitDiv(HDiv* div) OVERRIDE {
     57     BinaryFP(div);
     58   }
     59 
     60   void VisitCompare(HCompare* compare) OVERRIDE {
     61     BinaryFP(compare);
     62   }
     63 
     64   void VisitReturn(HReturn* ret) OVERRIDE {
     65     HConstant* value = ret->InputAt(0)->AsConstant();
     66     if ((value != nullptr && Primitive::IsFloatingPointType(value->GetType()))) {
     67       ReplaceInput(ret, value, 0, true);
     68     }
     69   }
     70 
     71   void VisitInvokeStaticOrDirect(HInvokeStaticOrDirect* invoke) OVERRIDE {
     72     HandleInvoke(invoke);
     73   }
     74 
     75   void VisitInvokeVirtual(HInvokeVirtual* invoke) OVERRIDE {
     76     HandleInvoke(invoke);
     77   }
     78 
     79   void VisitInvokeInterface(HInvokeInterface* invoke) OVERRIDE {
     80     HandleInvoke(invoke);
     81   }
     82 
     83   void VisitLoadClass(HLoadClass* load_class) OVERRIDE {
     84     HLoadClass::LoadKind load_kind = load_class->GetLoadKind();
     85     if (load_kind == HLoadClass::LoadKind::kBootImageLinkTimePcRelative ||
     86         load_kind == HLoadClass::LoadKind::kBssEntry) {
     87       HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(load_class);
     88       load_class->AddSpecialInput(method_address);
     89     }
     90   }
     91 
     92   void VisitLoadString(HLoadString* load_string) OVERRIDE {
     93     HLoadString::LoadKind load_kind = load_string->GetLoadKind();
     94     if (load_kind == HLoadString::LoadKind::kBootImageLinkTimePcRelative ||
     95         load_kind == HLoadString::LoadKind::kBssEntry) {
     96       HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(load_string);
     97       load_string->AddSpecialInput(method_address);
     98     }
     99   }
    100 
    101   void BinaryFP(HBinaryOperation* bin) {
    102     HConstant* rhs = bin->InputAt(1)->AsConstant();
    103     if (rhs != nullptr && Primitive::IsFloatingPointType(rhs->GetType())) {
    104       ReplaceInput(bin, rhs, 1, false);
    105     }
    106   }
    107 
    108   void VisitEqual(HEqual* cond) OVERRIDE {
    109     BinaryFP(cond);
    110   }
    111 
    112   void VisitNotEqual(HNotEqual* cond) OVERRIDE {
    113     BinaryFP(cond);
    114   }
    115 
    116   void VisitLessThan(HLessThan* cond) OVERRIDE {
    117     BinaryFP(cond);
    118   }
    119 
    120   void VisitLessThanOrEqual(HLessThanOrEqual* cond) OVERRIDE {
    121     BinaryFP(cond);
    122   }
    123 
    124   void VisitGreaterThan(HGreaterThan* cond) OVERRIDE {
    125     BinaryFP(cond);
    126   }
    127 
    128   void VisitGreaterThanOrEqual(HGreaterThanOrEqual* cond) OVERRIDE {
    129     BinaryFP(cond);
    130   }
    131 
    132   void VisitNeg(HNeg* neg) OVERRIDE {
    133     if (Primitive::IsFloatingPointType(neg->GetType())) {
    134       // We need to replace the HNeg with a HX86FPNeg in order to address the constant area.
    135       HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(neg);
    136       HGraph* graph = GetGraph();
    137       HBasicBlock* block = neg->GetBlock();
    138       HX86FPNeg* x86_fp_neg = new (graph->GetArena()) HX86FPNeg(
    139           neg->GetType(),
    140           neg->InputAt(0),
    141           method_address,
    142           neg->GetDexPc());
    143       block->ReplaceAndRemoveInstructionWith(neg, x86_fp_neg);
    144     }
    145   }
    146 
    147   void VisitPackedSwitch(HPackedSwitch* switch_insn) OVERRIDE {
    148     if (switch_insn->GetNumEntries() <=
    149         InstructionCodeGeneratorX86::kPackedSwitchJumpTableThreshold) {
    150       return;
    151     }
    152     // We need to replace the HPackedSwitch with a HX86PackedSwitch in order to
    153     // address the constant area.
    154     HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(switch_insn);
    155     HGraph* graph = GetGraph();
    156     HBasicBlock* block = switch_insn->GetBlock();
    157     HX86PackedSwitch* x86_switch = new (graph->GetArena()) HX86PackedSwitch(
    158         switch_insn->GetStartValue(),
    159         switch_insn->GetNumEntries(),
    160         switch_insn->InputAt(0),
    161         method_address,
    162         switch_insn->GetDexPc());
    163     block->ReplaceAndRemoveInstructionWith(switch_insn, x86_switch);
    164   }
    165 
    166   HX86ComputeBaseMethodAddress* GetPCRelativeBasePointer(HInstruction* cursor) {
    167     bool has_irreducible_loops = GetGraph()->HasIrreducibleLoops();
    168     if (!has_irreducible_loops) {
    169       // Ensure we only initialize the pointer once.
    170       if (base_ != nullptr) {
    171         return base_;
    172       }
    173     }
    174     // Insert the base at the start of the entry block, move it to a better
    175     // position later in MoveBaseIfNeeded().
    176     HX86ComputeBaseMethodAddress* method_address =
    177         new (GetGraph()->GetArena()) HX86ComputeBaseMethodAddress();
    178     if (has_irreducible_loops) {
    179       cursor->GetBlock()->InsertInstructionBefore(method_address, cursor);
    180     } else {
    181       HBasicBlock* entry_block = GetGraph()->GetEntryBlock();
    182       entry_block->InsertInstructionBefore(method_address, entry_block->GetFirstInstruction());
    183       base_ = method_address;
    184     }
    185     return method_address;
    186   }
    187 
    188   void ReplaceInput(HInstruction* insn, HConstant* value, int input_index, bool materialize) {
    189     HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(insn);
    190     HX86LoadFromConstantTable* load_constant =
    191         new (GetGraph()->GetArena()) HX86LoadFromConstantTable(method_address, value);
    192     if (!materialize) {
    193       load_constant->MarkEmittedAtUseSite();
    194     }
    195     insn->GetBlock()->InsertInstructionBefore(load_constant, insn);
    196     insn->ReplaceInput(load_constant, input_index);
    197   }
    198 
    199   void HandleInvoke(HInvoke* invoke) {
    200     // If this is an invoke-static/-direct with PC-relative dex cache array
    201     // addressing, we need the PC-relative address base.
    202     HInvokeStaticOrDirect* invoke_static_or_direct = invoke->AsInvokeStaticOrDirect();
    203     // We can't add a pointer to the constant area if we already have a current
    204     // method pointer. This may arise when sharpening doesn't remove the current
    205     // method pointer from the invoke.
    206     if (invoke_static_or_direct != nullptr &&
    207         invoke_static_or_direct->HasCurrentMethodInput()) {
    208       DCHECK(!invoke_static_or_direct->HasPcRelativeMethodLoadKind());
    209       return;
    210     }
    211 
    212     bool base_added = false;
    213     if (invoke_static_or_direct != nullptr &&
    214         invoke_static_or_direct->HasPcRelativeMethodLoadKind() &&
    215         !IsCallFreeIntrinsic<IntrinsicLocationsBuilderX86>(invoke, codegen_)) {
    216       HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(invoke);
    217       // Add the extra parameter.
    218       invoke_static_or_direct->AddSpecialInput(method_address);
    219       base_added = true;
    220     }
    221 
    222     // Ensure that we can load FP arguments from the constant area.
    223     HInputsRef inputs = invoke->GetInputs();
    224     for (size_t i = 0; i < inputs.size(); i++) {
    225       HConstant* input = inputs[i]->AsConstant();
    226       if (input != nullptr && Primitive::IsFloatingPointType(input->GetType())) {
    227         ReplaceInput(invoke, input, i, true);
    228       }
    229     }
    230 
    231     // These intrinsics need the constant area.
    232     switch (invoke->GetIntrinsic()) {
    233       case Intrinsics::kMathAbsDouble:
    234       case Intrinsics::kMathAbsFloat:
    235       case Intrinsics::kMathMaxDoubleDouble:
    236       case Intrinsics::kMathMaxFloatFloat:
    237       case Intrinsics::kMathMinDoubleDouble:
    238       case Intrinsics::kMathMinFloatFloat:
    239       case Intrinsics::kMathRoundFloat:
    240         if (!base_added) {
    241           DCHECK(invoke_static_or_direct != nullptr);
    242           DCHECK(!invoke_static_or_direct->HasCurrentMethodInput());
    243           HX86ComputeBaseMethodAddress* method_address = GetPCRelativeBasePointer(invoke);
    244           invoke_static_or_direct->AddSpecialInput(method_address);
    245         }
    246         break;
    247       default:
    248         break;
    249     }
    250   }
    251 
    252   CodeGeneratorX86* codegen_;
    253 
    254   // The generated HX86ComputeBaseMethodAddress in the entry block needed as an
    255   // input to the HX86LoadFromConstantTable instructions. Only set for
    256   // graphs with reducible loops.
    257   HX86ComputeBaseMethodAddress* base_;
    258 };
    259 
    260 void PcRelativeFixups::Run() {
    261   PCRelativeHandlerVisitor visitor(graph_, codegen_);
    262   visitor.VisitInsertionOrder();
    263   visitor.MoveBaseIfNeeded();
    264 }
    265 
    266 }  // namespace x86
    267 }  // namespace art
    268