diff options
| -rw-r--r-- | llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp | 30 | ||||
| -rw-r--r-- | llvm/test/CodeGen/ARM/flag-crash.ll | 27 |
2 files changed, 46 insertions, 11 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp b/llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp index 07277b3f349..c54f36002ea 100644 --- a/llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp +++ b/llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp @@ -110,18 +110,25 @@ static void CheckForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op, static void AddFlags(SDNode *N, SDValue Flag, bool AddFlag, SelectionDAG *DAG) { SmallVector<EVT, 4> VTs; + SDNode *FlagDestNode = Flag.getNode(); - for (unsigned i = 0, e = N->getNumValues(); i != e; ++i) - VTs.push_back(N->getValueType(i)); + // Don't add a flag from a node to itself. + if (FlagDestNode == N) return; + + // Don't add a flag to something which already has a flag. + if (N->getValueType(N->getNumValues() - 1) == MVT::Flag) return; + + for (unsigned I = 0, E = N->getNumValues(); I != E; ++I) + VTs.push_back(N->getValueType(I)); if (AddFlag) VTs.push_back(MVT::Flag); SmallVector<SDValue, 4> Ops; - for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) - Ops.push_back(N->getOperand(i)); + for (unsigned I = 0, E = N->getNumOperands(); I != E; ++I) + Ops.push_back(N->getOperand(I)); - if (Flag.getNode()) + if (FlagDestNode) Ops.push_back(Flag); SDVTList VTList = DAG->getVTList(&VTs[0], VTs.size()); @@ -213,16 +220,17 @@ void ScheduleDAGSDNodes::ClusterNeighboringLoads(SDNode *Node) { // Cluster loads by adding MVT::Flag outputs and inputs. This also // ensure they are scheduled in order of increasing addresses. SDNode *Lead = Loads[0]; - AddFlags(Lead, SDValue(0,0), true, DAG); + AddFlags(Lead, SDValue(0, 0), true, DAG); + + SDValue InFlag = SDValue(Lead, Lead->getNumValues() - 1); + for (unsigned I = 1, E = Loads.size(); I != E; ++I) { + bool OutFlag = I < E - 1; + SDNode *Load = Loads[I]; - SDValue InFlag = SDValue(Lead, Lead->getNumValues()-1); - for (unsigned i = 1, e = Loads.size(); i != e; ++i) { - bool OutFlag = i < e-1; - SDNode *Load = Loads[i]; AddFlags(Load, InFlag, OutFlag, DAG); if (OutFlag) - InFlag = SDValue(Load, Load->getNumValues()-1); + InFlag = SDValue(Load, Load->getNumValues() - 1); ++LoadsClustered; } diff --git a/llvm/test/CodeGen/ARM/flag-crash.ll b/llvm/test/CodeGen/ARM/flag-crash.ll new file mode 100644 index 00000000000..9c61944a215 --- /dev/null +++ b/llvm/test/CodeGen/ARM/flag-crash.ll @@ -0,0 +1,27 @@ +; RUN: llc < %s -O3 -mtriple=thumbv7-apple-darwin10 -mcpu=cortex-a8 -relocation-model=pic +; PR7484 + +%struct.gs_matrix = type { float, i32, float, i32, float, i32, float, i32, float, i32, float, i32 } + +define fastcc void @func(%struct.gs_matrix* nocapture %pm1) nounwind { +entry: + %0 = getelementptr inbounds %struct.gs_matrix* %pm1, i32 0, i32 6 + %1 = load float* %0, align 4 + %2 = getelementptr inbounds %struct.gs_matrix* %pm1, i32 0, i32 8 + %3 = load float* %2, align 4 + %4 = getelementptr inbounds %struct.gs_matrix* %pm1, i32 0, i32 2 + %5 = bitcast float* %4 to i32* + %6 = load i32* %5, align 4 + %7 = or i32 0, %6 + %.mask = and i32 %7, 2147483647 + %8 = icmp eq i32 %.mask, 0 + br i1 %8, label %bb, label %bb11 + +bb: + ret void + +bb11: + %9 = fmul float %1, undef + %10 = fmul float %3, undef + ret void +} |
