OLD | NEW |
---|---|
1 //===- subzero/src/IceCfg.cpp - Control flow graph implementation ---------===// | 1 //===- subzero/src/IceCfg.cpp - Control flow graph implementation ---------===// |
2 // | 2 // |
3 // The Subzero Code Generator | 3 // The Subzero Code Generator |
4 // | 4 // |
5 // This file is distributed under the University of Illinois Open Source | 5 // This file is distributed under the University of Illinois Open Source |
6 // License. See LICENSE.TXT for details. | 6 // License. See LICENSE.TXT for details. |
7 // | 7 // |
8 //===----------------------------------------------------------------------===// | 8 //===----------------------------------------------------------------------===// |
9 /// | 9 /// |
10 /// \file | 10 /// \file |
(...skipping 770 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
781 } | 781 } |
782 } | 782 } |
783 | 783 |
784 SizeT NodeIndex = 0; | 784 SizeT NodeIndex = 0; |
785 for (auto *Node : NewList) { | 785 for (auto *Node : NewList) { |
786 Node->resetIndex(NodeIndex++); | 786 Node->resetIndex(NodeIndex++); |
787 } | 787 } |
788 Nodes = NewList; | 788 Nodes = NewList; |
789 } | 789 } |
790 | 790 |
791 namespace { | |
792 | |
793 /// VariableMap is a simple helper class for splitLocalVars(), that keeps track | |
794 /// of the latest split version of the original Variables. For each entry, the | |
795 /// Variable is tagged with the CfgNode that it is valid in, so that we don't | |
796 /// need to clear the entire Map[] vector for each block. | |
797 class VariableMap { | |
798 private: | |
799 VariableMap() = delete; | |
800 VariableMap(const VariableMap &) = delete; | |
801 VariableMap &operator=(const VariableMap &) = delete; | |
802 | |
803 /// VarNodePair is basically std::pair<Variable*,CfgNode*> . | |
804 struct VarNodePair { | |
805 Variable *Var = nullptr; | |
806 const CfgNode *Node = nullptr; | |
807 VarNodePair() = default; | |
808 | |
809 private: | |
810 VarNodePair(const VarNodePair &) = delete; | |
811 VarNodePair &operator=(const VarNodePair &) = delete; | |
812 }; | |
813 | |
814 public: | |
815 explicit VariableMap(Cfg *Func) | |
816 : Func(Func), NumVars(Func->getNumVariables()), Map(NumVars) {} | |
817 /// Reset the mappings at the start of a block. | |
818 void reset(const CfgNode *CurNode) { Node = CurNode; } | |
819 /// Get Var's current mapping (or Var itself if it has no mapping yet). | |
820 Variable *get(Variable *Var) const { | |
821 const SizeT VarNum = getVarNum(Var); | |
822 Variable *MappedVar = Map[VarNum].Var; | |
823 if (MappedVar == nullptr) | |
824 return Var; | |
825 if (Map[VarNum].Node != Node) | |
826 return Var; | |
827 return MappedVar; | |
828 } | |
829 /// Create a new linked Variable in the LinkedTo chain, and set it as Var's | |
830 /// latest mapping. | |
831 Variable *makeLinked(Variable *Var) { | |
832 Variable *NewVar = Func->makeVariable(Var->getType()); | |
833 NewVar->setRegClass(Var->getRegClass()); | |
834 NewVar->setLinkedTo(get(Var)); | |
835 const SizeT VarNum = getVarNum(Var); | |
836 Map[VarNum].Var = NewVar; | |
837 Map[VarNum].Node = Node; | |
838 return NewVar; | |
839 } | |
840 /// Given Var that is LinkedTo some other variable, re-splice it into the | |
841 /// LinkedTo chain so that the chain is ordered by Variable::getIndex(). | |
842 void spliceBlockLocalLinkedToChain(Variable *Var) { | |
843 Variable *LinkedTo = Var->getLinkedTo(); | |
844 assert(LinkedTo != nullptr); | |
845 assert(Var->getIndex() > LinkedTo->getIndex()); | |
846 const SizeT VarNum = getVarNum(LinkedTo); | |
847 Variable *Link = Map[VarNum].Var; | |
848 if (Link == nullptr || Map[VarNum].Node != Node) | |
849 return; | |
850 Variable *LinkParent = Link->getLinkedTo(); | |
851 while (LinkParent != nullptr && LinkParent->getIndex() >= Var->getIndex()) { | |
852 Link = LinkParent; | |
853 LinkParent = Link->getLinkedTo(); | |
854 } | |
855 Var->setLinkedTo(LinkParent); | |
856 Link->setLinkedTo(Var); | |
857 } | |
858 | |
859 private: | |
860 Cfg *const Func; | |
861 // NumVars is for the size of the Map array. It can be const because any new | |
862 // Variables created during the splitting pass don't need to be mapped. | |
863 const SizeT NumVars; | |
864 CfgVector<VarNodePair> Map; | |
865 const CfgNode *Node = nullptr; | |
866 /// Get Var's VarNum, and do some validation. | |
867 SizeT getVarNum(Variable *Var) const { | |
868 const SizeT VarNum = Var->getIndex(); | |
869 assert(VarNum < NumVars); | |
870 return VarNum; | |
871 } | |
872 }; | |
873 | |
874 /// A Variable is "allocable" if it is a register allocation candidate but | |
875 /// doesn't already have a register. | |
876 bool isAllocable(const Variable *Var) { | |
877 if (Var == nullptr) | |
878 return false; | |
879 return !Var->hasReg() && Var->mayHaveReg(); | |
880 } | |
881 | |
882 /// A Variable is "inf" if it already has a register or is infinite-weight. | |
883 bool isInf(const Variable *Var) { | |
884 if (Var == nullptr) | |
885 return false; | |
886 return Var->hasReg() || Var->mustHaveReg(); | |
887 } | |
888 | |
889 } // end of anonymous namespace | |
890 | |
891 /// Within each basic block, rewrite Variable references in terms of chained | |
892 /// copies of the original Variable. For example: | |
893 /// A = B + C | |
894 /// might be rewritten as: | |
895 /// B1 = B | |
896 /// C1 = C | |
897 /// A = B + C | |
898 /// A1 = A | |
899 /// and then: | |
900 /// D = A + B | |
901 /// might be rewritten as: | |
902 /// A2 = A1 | |
903 /// B2 = B1 | |
904 /// D = A1 + B1 | |
905 /// D1 = D | |
906 /// | |
907 /// The purpose is to present the linear-scan register allocator with smaller | |
908 /// live ranges, to help mitigate its "all or nothing" allocation strategy, | |
909 /// while counting on its preference mechanism to keep the split versions in the | |
910 /// same register when possible. | |
911 /// | |
912 /// When creating new Variables, A2 is linked to A1 which is linked to A, and | |
913 /// similar for the other Variable linked-to chains. Rewrites apply only to | |
914 /// Variables where mayHaveReg() is true. | |
915 /// | |
916 /// At code emission time, redundant linked-to stack assignments will be | |
917 /// recognized and elided. To illustrate using the above example, if A1 gets a | |
918 /// register but A and A2 are on the stack, the "A2=A1" store instruction is | |
919 /// redundant since A and A2 share the same stack slot and A1 originated from A. | |
920 /// | |
921 /// Simple assignment instructions are rewritten slightly differently, to take | |
922 /// maximal advantage of Variables known to have registers. | |
923 /// | |
924 /// In general, there may be several valid ways to rewrite an instruction: add | |
925 /// the new assignment instruction either before or after the original | |
926 /// instruction, and rewrite the original instruction with either the old or the | |
927 /// new variable mapping. We try to pick a strategy most likely to avoid | |
928 /// potential performance problems. For example, try to avoid storing to the | |
929 /// stack and then immediately reloading from the same location. One | |
930 /// consequence is that code might be generated that loads a register from a | |
931 /// stack location, followed almost immediately by another use of the same stack | |
932 /// location, despite its value already being available in a register as a | |
933 /// result of the first instruction. However, the performance impact here is | |
934 /// likely to be negligible, and a simple availability peephole optimization | |
935 /// could clean it up. | |
936 /// | |
937 /// This pass potentially adds a lot of new instructions and variables, and as | |
938 /// such there are compile-time performance concerns, particularly with liveness | |
939 /// analysis and register allocation. Note that for liveness analysis, the new | |
940 /// variables have single-block liveness, so they don't increase the size of the | |
941 /// liveness bit vectors that need to be merged across blocks. As a result, the | |
942 /// performance impact is likely to be linearly related to the number of new | |
943 /// instructions, rather than number of new variables times number of blocks | |
944 /// which would be the case if they were multi-block variables. | |
945 void Cfg::splitLocalVars() { | |
946 if (!getFlags().getSplitLocalVars()) | |
947 return; | |
948 TimerMarker _(TimerStack::TT_splitLocalVars, this); | |
949 VariableMap VarMap(this); | |
950 // TODO(stichnot): Fix this mechanism for LinkedTo variables and stack slot | |
951 // assignment. | |
952 // | |
953 // To work around shortcomings with stack frame mapping, we want to arrange | |
954 // LinkedTo structure such that within one block, the LinkedTo structure | |
955 // leading to a root forms a list, not a tree. A LinkedTo root can have | |
956 // multiple children linking to it, but only one per block. Furthermore, | |
957 // because stack slot mapping processes variables in numerical order, the | |
958 // LinkedTo chain needs to be ordered such that when A->getLinkedTo()==B, then | |
959 // A->getIndex()>B->getIndex(). | |
John
2016/08/01 14:17:31
add spaces around ' > ' (the gt operator)
Jim Stichnoth
2016/08/01 15:13:50
Done.
| |
960 // | |
961 // To effect this, while processing a block we keep track of preexisting | |
962 // LinkedTo relationships via the LinkedToFixups vector, and at the end of the | |
963 // block we splice them in such that the block has a single chain for each | |
964 // root, ordered by getIndex() value. | |
965 CfgVector<Variable *> LinkedToFixups; | |
966 for (CfgNode *Node : getNodes()) { | |
967 // Clear the VarMap and LinkedToFixups at the start of every block. | |
968 VarMap.reset(Node); | |
969 LinkedToFixups.clear(); | |
970 auto &Insts = Node->getInsts(); | |
971 auto Iter = Insts.begin(); | |
972 auto IterEnd = Insts.end(); | |
973 // TODO(stichnot): Also create assignments/mappings for phi dest variables. | |
974 InstList::iterator NextIter; | |
975 const Inst *WaitingForLabel = nullptr; | |
976 const Inst *WaitingForBranchTo = nullptr; | |
977 for (; Iter != IterEnd; Iter = NextIter) { | |
978 NextIter = Iter; | |
979 ++NextIter; | |
980 Inst *Instr = iteratorToInst(Iter); | |
981 if (Instr->isDeleted()) | |
982 continue; | |
983 Variable *Dest = Instr->getDest(); | |
984 | |
985 // Before doing any transformations, take care of the bookkeeping for | |
986 // intra-block branching. | |
987 // | |
988 // This is tricky because the transformation for one instruction may | |
989 // depend on a transformation for a previous instruction, but if that | |
990 // previous instruction is not dynamically executed due to intra-block | |
991 // control flow, it may lead to an inconsistent state and incorrect code. | |
992 // | |
993 // We want to handle some simple cases, and reject some others: | |
994 // | |
995 // 1. For something like a select instruction, we could have: | |
996 // test cond | |
997 // dest = src_false | |
998 // branch conditionally to label | |
999 // dest = src_true | |
1000 // label: | |
1001 // | |
1002 // Between the conditional branch and the label, we need to treat dest and | |
1003 // src variables specially, specifically not creating any new state. | |
1004 // | |
1005 // 2. Some 64-bit atomic instructions may be lowered to a loop: | |
1006 // label: | |
1007 // ... | |
1008 // branch conditionally to label | |
1009 // | |
1010 // No special treatment is needed, but it's worth tracking so that case #1 | |
1011 // above can also be handled. | |
1012 // | |
1013 // 3. Advanced switch lowering can create really complex intra-block | |
1014 // control flow, so when we recognize this, we should just stop splitting | |
1015 // for the remainder of the block (which isn't much since a switch | |
1016 // instruction is a terminator). | |
1017 // | |
1018 // 4. Other complex lowering, e.g. an i64 icmp on a 32-bit architecture, | |
1019 // can result in an if/then/else like structure with two labels. One | |
1020 // possibility would be to suspect splitting for the remainder of the | |
1021 // lowered instruction, and then resume for the remainder of the block, | |
1022 // but since we don't have high-level instruction markers, we might as | |
1023 // well just stop splitting for the remainder of the block. | |
1024 if (Instr->isLabel()) { | |
1025 // A Label instruction shouldn't have any operands, so it can be handled | |
1026 // right here and then move on. | |
1027 assert(Dest == nullptr); | |
1028 assert(Instr->getSrcSize() == 0); | |
1029 if (Instr == WaitingForLabel) { | |
1030 // If we found the forward-branch-target Label instruction we're | |
1031 // waiting for, then clear the WaitingForLabel state. | |
1032 WaitingForLabel = nullptr; | |
1033 } else if (WaitingForLabel == nullptr && | |
1034 WaitingForBranchTo == nullptr) { | |
1035 // If we found a new Label instruction while the WaitingFor* state is | |
1036 // clear, then set things up for this being a backward branch target. | |
1037 WaitingForBranchTo = Instr; | |
1038 } else { | |
1039 // We see something we don't understand, so skip to the next block. | |
1040 break; | |
1041 } | |
1042 continue; // move to next instruction | |
1043 } | |
1044 if (const Inst *Label = Instr->getIntraBlockBranchTarget()) { | |
1045 // An intra-block branch instruction shouldn't have any operands, so it | |
1046 // can be handled right here and then move on. | |
1047 assert(Dest == nullptr); | |
1048 assert(Instr->getSrcSize() == 0); | |
1049 if (WaitingForBranchTo == Label && WaitingForLabel == nullptr) { | |
1050 WaitingForBranchTo = nullptr; | |
1051 } else if (WaitingForBranchTo == nullptr && | |
1052 (WaitingForLabel == nullptr || WaitingForLabel == Label)) { | |
1053 WaitingForLabel = Label; | |
1054 } else { | |
1055 // We see something we don't understand, so skip to the next block. | |
1056 break; | |
1057 } | |
1058 continue; // move to next instruction | |
1059 } | |
1060 | |
1061 // Intra-block bookkeeping is complete, now do the transformations. | |
1062 static constexpr char AnInstructionHasNoName[] = ""; | |
1063 // We can limit the splitting to an arbitrary subset of the instructions, | |
1064 // and still expect correct code. As such, we can do instruction-subset | |
1065 // bisection to help debug any problems in this pass. | |
1066 if (!BuildDefs::minimal() && | |
1067 !getFlags().matchSplitInsts(AnInstructionHasNoName, | |
1068 Instr->getNumber())) | |
1069 continue; | |
1070 | |
1071 if (!llvm::isa<InstTarget>(Instr)) { | |
1072 // Ignore non-lowered instructions like FakeDef/FakeUse. | |
1073 continue; | |
1074 } | |
1075 const bool IsUnconditionallyExecuted = (WaitingForLabel == nullptr); | |
1076 const bool DestIsInf = isInf(Dest); | |
1077 const bool DestIsAllocable = isAllocable(Dest); | |
1078 // Note any preexisting LinkedTo relationships that were created during | |
1079 // target lowering. | |
1080 if (Dest != nullptr && Dest->getLinkedTo() != nullptr) { | |
1081 LinkedToFixups.emplace_back(Dest); | |
John
2016/08/01 14:17:31
Any reasons why you can't spliceBlockLocalLinkedTo
Jim Stichnoth
2016/08/01 15:13:50
I no longer remember the exact scenario, but this
| |
1082 } | |
1083 // Determine the transformation based on the kind of instruction, and | |
1084 // whether its Variables are infinite-weight. New instructions can be | |
1085 // inserted before the current instruction via Iter, or after the current | |
1086 // instruction via NextIter. | |
1087 if (Instr->isVarAssign()) { | |
1088 auto *SrcVar = llvm::cast<Variable>(Instr->getSrc(0)); | |
1089 const bool SrcIsInf = isInf(SrcVar); | |
1090 const bool SrcIsAllocable = isAllocable(SrcVar); | |
1091 if (DestIsInf && SrcIsInf) { | |
1092 // The instruction: | |
1093 // t:inf = u:inf | |
1094 // No transformation is needed. | |
1095 continue; | |
1096 } else if (DestIsInf && SrcIsAllocable && | |
1097 Dest->getType() == Instr->getSrc(0)->getType()) { | |
1098 // The instruction: | |
1099 // t:inf = v | |
1100 // gets transformed to: | |
1101 // t:inf = v1 | |
1102 // v2 = t:inf | |
1103 // where: | |
1104 // v1 := map[v] | |
1105 // v2 := linkTo(v) | |
1106 // map[v] := v2 | |
1107 // | |
1108 // If both v2 and its linkedToStackRoot get a stack slot, then | |
1109 // "v2=t:inf" is recognized as a redundant assignment and elided. | |
1110 // | |
1111 // Note that if the dest and src types are different, then this is | |
1112 // actually a truncation operation, which would make "v2=t:inf" an | |
1113 // invalid instruction. In this case, the type test will make it fall | |
1114 // through to the general case below. | |
1115 Variable *OldMapped = VarMap.get(SrcVar); | |
1116 Instr->replaceSource(0, OldMapped); | |
1117 if (IsUnconditionallyExecuted) { | |
1118 // Only create new mapping state if the instruction is | |
1119 // unconditionally executed. | |
1120 Variable *NewMapped = VarMap.makeLinked(SrcVar); | |
1121 Inst *Mov = Target->createLoweredMove(NewMapped, Dest); | |
1122 Insts.insert(NextIter, Mov); | |
1123 } | |
1124 continue; | |
1125 } else if (DestIsAllocable && SrcIsInf) { | |
1126 // The instruction: | |
1127 // v = t:inf | |
1128 // gets transformed to: | |
1129 // v = t:inf | |
1130 // v2 = t:inf | |
1131 // where: | |
1132 // v2 := linkTo(v) | |
1133 // map[v] := v2 | |
1134 // | |
1135 // If both v2 and v get a stack slot, then "v2=t:inf" is recognized as | |
1136 // a redundant assignment and elided. | |
1137 if (IsUnconditionallyExecuted) { | |
1138 // Only create new mapping state if the instruction is | |
1139 // unconditionally executed. | |
1140 Variable *NewMapped = VarMap.makeLinked(Dest); | |
1141 Inst *Mov = Target->createLoweredMove(NewMapped, SrcVar); | |
1142 Insts.insert(NextIter, Mov); | |
1143 } else { | |
1144 // For a conditionally executed instruction, add a redefinition of | |
1145 // the original Dest mapping, without creating a new linked | |
1146 // variable. | |
1147 Variable *OldMapped = VarMap.get(Dest); | |
1148 Inst *Mov = Target->createLoweredMove(OldMapped, SrcVar); | |
1149 Mov->setDestRedefined(); | |
1150 Insts.insert(NextIter, Mov); | |
1151 } | |
1152 continue; | |
1153 } | |
1154 } | |
1155 // The (non-variable-assignment) instruction: | |
1156 // ... = F(v) | |
1157 // where v is not infinite-weight, gets transformed to: | |
1158 // v2 = v1 | |
1159 // ... = F(v1) | |
1160 // where: | |
1161 // v1 := map[v] | |
1162 // v2 := linkTo(v) | |
1163 // map[v] := v2 | |
1164 // After that, if the "..." dest=u is not infinite-weight, append: | |
1165 // u2 = u | |
1166 // where: | |
1167 // u2 := linkTo(u) | |
1168 // map[u] := u2 | |
1169 for (SizeT i = 0; i < Instr->getSrcSize(); ++i) { | |
1170 // Iterate over the top-level src vars. Don't bother to dig into | |
1171 // e.g. MemOperands because their vars should all be infinite-weight. | |
1172 // (This assumption would need to change if the pass were done | |
1173 // pre-lowering.) | |
1174 if (auto *SrcVar = llvm::dyn_cast<Variable>(Instr->getSrc(i))) { | |
1175 const bool SrcIsAllocable = isAllocable(SrcVar); | |
1176 if (SrcIsAllocable) { | |
1177 Variable *OldMapped = VarMap.get(SrcVar); | |
1178 if (IsUnconditionallyExecuted) { | |
1179 Variable *NewMapped = VarMap.makeLinked(SrcVar); | |
1180 Inst *Mov = Target->createLoweredMove(NewMapped, OldMapped); | |
1181 Insts.insert(Iter, Mov); | |
1182 } | |
1183 Instr->replaceSource(i, OldMapped); | |
1184 } | |
1185 } | |
1186 } | |
1187 // Transformation of Dest is the same as the "v=t:inf" case above. | |
1188 if (DestIsAllocable) { | |
1189 if (IsUnconditionallyExecuted) { | |
1190 Variable *NewMapped = VarMap.makeLinked(Dest); | |
1191 Inst *Mov = Target->createLoweredMove(NewMapped, Dest); | |
1192 Insts.insert(NextIter, Mov); | |
1193 } else { | |
1194 Variable *OldMapped = VarMap.get(Dest); | |
1195 Inst *Mov = Target->createLoweredMove(OldMapped, Dest); | |
1196 Mov->setDestRedefined(); | |
1197 Insts.insert(NextIter, Mov); | |
1198 } | |
1199 } | |
1200 } | |
1201 // Splice in any preexisting links into the single chain. | |
1202 for (Variable *Var : LinkedToFixups) { | |
1203 VarMap.spliceBlockLocalLinkedToChain(Var); | |
1204 } | |
1205 } | |
1206 dump("After splitting local variables"); | |
1207 } | |
1208 | |
791 void Cfg::doArgLowering() { | 1209 void Cfg::doArgLowering() { |
792 TimerMarker T(TimerStack::TT_doArgLowering, this); | 1210 TimerMarker T(TimerStack::TT_doArgLowering, this); |
793 getTarget()->lowerArguments(); | 1211 getTarget()->lowerArguments(); |
794 } | 1212 } |
795 | 1213 |
796 void Cfg::sortAndCombineAllocas(CfgVector<InstAlloca *> &Allocas, | 1214 void Cfg::sortAndCombineAllocas(CfgVector<InstAlloca *> &Allocas, |
797 uint32_t CombinedAlignment, InstList &Insts, | 1215 uint32_t CombinedAlignment, InstList &Insts, |
798 AllocaBaseVariableType BaseVariableType) { | 1216 AllocaBaseVariableType BaseVariableType) { |
799 if (Allocas.empty()) | 1217 if (Allocas.empty()) |
800 return; | 1218 return; |
(...skipping 853 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
1654 dump("After recomputing liveness for -decorate-asm"); | 2072 dump("After recomputing liveness for -decorate-asm"); |
1655 } | 2073 } |
1656 OstreamLocker L(Ctx); | 2074 OstreamLocker L(Ctx); |
1657 Ostream &Str = Ctx->getStrEmit(); | 2075 Ostream &Str = Ctx->getStrEmit(); |
1658 const Assembler *Asm = getAssembler<>(); | 2076 const Assembler *Asm = getAssembler<>(); |
1659 const bool NeedSandboxing = getFlags().getUseSandboxing(); | 2077 const bool NeedSandboxing = getFlags().getUseSandboxing(); |
1660 | 2078 |
1661 emitTextHeader(FunctionName, Ctx, Asm); | 2079 emitTextHeader(FunctionName, Ctx, Asm); |
1662 if (getFlags().getDecorateAsm()) { | 2080 if (getFlags().getDecorateAsm()) { |
1663 for (Variable *Var : getVariables()) { | 2081 for (Variable *Var : getVariables()) { |
1664 if (Var->hasStackOffset() && !Var->isRematerializable()) { | 2082 if (Var->hasKnownStackOffset() && !Var->isRematerializable()) { |
1665 Str << "\t" << Var->getSymbolicStackOffset() << " = " | 2083 Str << "\t" << Var->getSymbolicStackOffset() << " = " |
1666 << Var->getStackOffset() << "\n"; | 2084 << Var->getStackOffset() << "\n"; |
1667 } | 2085 } |
1668 } | 2086 } |
1669 } | 2087 } |
1670 for (CfgNode *Node : Nodes) { | 2088 for (CfgNode *Node : Nodes) { |
1671 if (NeedSandboxing && Node->needsAlignment()) { | 2089 if (NeedSandboxing && Node->needsAlignment()) { |
1672 Str << "\t" << Asm->getAlignDirective() << " " | 2090 Str << "\t" << Asm->getAlignDirective() << " " |
1673 << Asm->getBundleAlignLog2Bytes() << "\n"; | 2091 << Asm->getBundleAlignLog2Bytes() << "\n"; |
1674 } | 2092 } |
(...skipping 92 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... | |
1767 } | 2185 } |
1768 } | 2186 } |
1769 // Print each basic block | 2187 // Print each basic block |
1770 for (CfgNode *Node : Nodes) | 2188 for (CfgNode *Node : Nodes) |
1771 Node->dump(this); | 2189 Node->dump(this); |
1772 if (isVerbose(IceV_Instructions)) | 2190 if (isVerbose(IceV_Instructions)) |
1773 Str << "}\n"; | 2191 Str << "}\n"; |
1774 } | 2192 } |
1775 | 2193 |
1776 } // end of namespace Ice | 2194 } // end of namespace Ice |
OLD | NEW |