Fix a bug in dead spill slot elimination.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@47687 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/CodeGen/MachineFrameInfo.h b/include/llvm/CodeGen/MachineFrameInfo.h
index 4be3bd474e0c418e7a669e6e03006294c1680273..3a0432510c86d851c2a01b1ff54f6820a02b9221 100644 (file)
--- a/include/llvm/CodeGen/MachineFrameInfo.h
+++ b/include/llvm/CodeGen/MachineFrameInfo.h
@@ -216,6 +216,8 @@ public:
   int64_t getObjectOffset(int ObjectIdx) const {
     assert(unsigned(ObjectIdx+NumFixedObjects) < Objects.size() &&
            "Invalid Object Idx!");
+    assert(!isDeadObjectIndex(ObjectIdx) &&
+           "Getting frame offset for a dead object?");
     return Objects[ObjectIdx+NumFixedObjects].SPOffset;
   }
 
@@ -225,6 +227,8 @@ public:
   void setObjectOffset(int ObjectIdx, int64_t SPOffset) {
     assert(unsigned(ObjectIdx+NumFixedObjects) < Objects.size() &&
            "Invalid Object Idx!");
+    assert(!isDeadObjectIndex(ObjectIdx) &&
+           "Setting frame offset for a dead object?");
     Objects[ObjectIdx+NumFixedObjects].SPOffset = SPOffset;
   }
 

diff --git a/lib/CodeGen/VirtRegMap.cpp b/lib/CodeGen/VirtRegMap.cpp
index d210d93835d0ca14c8f5e180254975a15045d4a0..6cfbee9b7b73f5c6e3846a0645eac62da607c5ad 100644 (file)
--- a/lib/CodeGen/VirtRegMap.cpp
+++ b/lib/CodeGen/VirtRegMap.cpp
@@ -914,6 +914,7 @@ bool LocalSpiller::PrepForUnfoldOpti(MachineBasicBlock &MBB,
       Ops.push_back(Idx);
       MachineInstr *FoldedMI = TII->foldMemoryOperand(MF, NewMI, Ops, SS);
       if (FoldedMI) {
+        VRM.addSpillSlotUse(SS, FoldedMI);
         if (!VRM.hasPhys(UnfoldVR))
           VRM.assignVirt2Phys(UnfoldVR, UnfoldPR);
         VRM.virtFolded(VirtReg, FoldedMI, VirtRegMap::isRef);
@@ -1414,6 +1415,7 @@ void LocalSpiller::RewriteMBB(MachineBasicBlock &MBB, VirtRegMap &VRM) {
             MBB.insert(MII, NewMIs[0]);
             NewStore = NewMIs[1];
             MBB.insert(MII, NewStore);
+            VRM.addSpillSlotUse(SS, NewStore);
             VRM.RemoveMachineInstrFromMaps(&MI);
             MBB.erase(&MI);
             Erased = true;

diff --git a/test/CodeGen/X86/2008-02-27-DeadSlotElimBug.ll b/test/CodeGen/X86/2008-02-27-DeadSlotElimBug.ll
new file mode 100644 (file)
index 0000000..96ac7a7
--- /dev/null
+++ b/test/CodeGen/X86/2008-02-27-DeadSlotElimBug.ll
@@ -0,0 +1,66 @@
+; RUN: llvm-as < %s | llc -march=x86
+
+       %struct.CompAtom = type <{ %struct.Position, float, i32 }>
+       %struct.Lattice = type { %struct.Position, %struct.Position, %struct.Position, %struct.Position, %struct.Position, %struct.Position, %struct.Position, i32, i32, i32 }
+       %struct.Position = type { double, double, double }
+
+define fastcc %struct.CompAtom* @_ZNK7Lattice6createEP8CompAtomii(%struct.Lattice* %this, %struct.CompAtom* %d, i32 %n, i32 %i) {
+entry:
+       %tmp18 = tail call i8* @_Znam( i32 0 )          ; <i8*> [#uses=1]
+       %tmp1819 = bitcast i8* %tmp18 to %struct.CompAtom*              ; <%struct.CompAtom*> [#uses=4]
+       %tmp3286 = icmp eq i32 %n, 0            ; <i1> [#uses=1]
+       br i1 %tmp3286, label %bb35, label %bb24
+
+bb24:          ; preds = %bb24, %entry
+       %tmp9.0.reg2mem.0.rec = phi i32 [ %indvar.next, %bb24 ], [ 0, %entry ]          ; <i32> [#uses=3]
+       %tmp3.i.i = getelementptr %struct.CompAtom* %tmp1819, i32 %tmp9.0.reg2mem.0.rec, i32 0, i32 1           ; <double*> [#uses=0]
+       %tmp5.i.i = getelementptr %struct.CompAtom* %tmp1819, i32 %tmp9.0.reg2mem.0.rec, i32 0, i32 2           ; <double*> [#uses=1]
+       store double -9.999900e+04, double* %tmp5.i.i, align 4
+       %indvar.next = add i32 %tmp9.0.reg2mem.0.rec, 1         ; <i32> [#uses=2]
+       %exitcond = icmp eq i32 %indvar.next, %n                ; <i1> [#uses=1]
+       br i1 %exitcond, label %bb35, label %bb24
+
+bb35:          ; preds = %bb24, %entry
+       %tmp42 = sdiv i32 %i, 9         ; <i32> [#uses=1]
+       %tmp43 = add i32 %tmp42, -1             ; <i32> [#uses=1]
+       %tmp4344 = sitofp i32 %tmp43 to double          ; <double> [#uses=1]
+       %tmp17.i76 = mul double %tmp4344, 0.000000e+00          ; <double> [#uses=1]
+       %tmp48 = sdiv i32 %i, 3         ; <i32> [#uses=1]
+       %tmp49 = srem i32 %tmp48, 3             ; <i32> [#uses=1]
+       %tmp50 = add i32 %tmp49, -1             ; <i32> [#uses=1]
+       %tmp5051 = sitofp i32 %tmp50 to double          ; <double> [#uses=1]
+       %tmp17.i63 = mul double %tmp5051, 0.000000e+00          ; <double> [#uses=1]
+       %tmp55 = srem i32 %i, 3         ; <i32> [#uses=1]
+       %tmp56 = add i32 %tmp55, -1             ; <i32> [#uses=1]
+       %tmp5657 = sitofp i32 %tmp56 to double          ; <double> [#uses=1]
+       %tmp15.i49 = getelementptr %struct.Lattice* %this, i32 0, i32 0, i32 0          ; <double*> [#uses=1]
+       %tmp16.i50 = load double* %tmp15.i49, align 4           ; <double> [#uses=1]
+       %tmp17.i = mul double %tmp5657, %tmp16.i50              ; <double> [#uses=1]
+       %tmp20.i39 = add double %tmp17.i, %tmp17.i63            ; <double> [#uses=1]
+       %tmp20.i23 = add double %tmp20.i39, %tmp17.i76          ; <double> [#uses=1]
+       br i1 false, label %bb58.preheader, label %bb81
+
+bb58.preheader:                ; preds = %bb35
+       %smax = select i1 false, i32 1, i32 %n          ; <i32> [#uses=1]
+       br label %bb58
+
+bb58:          ; preds = %bb58, %bb58.preheader
+       %tmp20.i7 = getelementptr %struct.CompAtom* %d, i32 0, i32 2            ; <i32*> [#uses=2]
+       %tmp25.i = getelementptr %struct.CompAtom* %tmp1819, i32 0, i32 2               ; <i32*> [#uses=2]
+       %tmp74.i = load i32* %tmp20.i7, align 1         ; <i32> [#uses=1]
+       %tmp82.i = and i32 %tmp74.i, 134217728          ; <i32> [#uses=1]
+       %tmp85.i = or i32 0, %tmp82.i           ; <i32> [#uses=1]
+       store i32 %tmp85.i, i32* %tmp25.i, align 1
+       %tmp88.i = load i32* %tmp20.i7, align 1         ; <i32> [#uses=1]
+       %tmp95.i = and i32 %tmp88.i, -268435456         ; <i32> [#uses=1]
+       %tmp97.i = or i32 0, %tmp95.i           ; <i32> [#uses=1]
+       store i32 %tmp97.i, i32* %tmp25.i, align 1
+       %tmp6.i = add double 0.000000e+00, %tmp20.i23           ; <double> [#uses=0]
+       %exitcond96 = icmp eq i32 0, %smax              ; <i1> [#uses=1]
+       br i1 %exitcond96, label %bb81, label %bb58
+
+bb81:          ; preds = %bb58, %bb35
+       ret %struct.CompAtom* %tmp1819
+}
+
+declare i8* @_Znam(i32)