After r147827 and r147902, it's now possible for unallocatable registers to be

live across BBs before register allocation. This miscompiled 197.parser
when a cmp + b are optimized to a cbnz instruction even though the CPSR def
is live-in a successor.
        cbnz    r6, LBB89_12
...
LBB89_12:
        ble     LBB89_1

The fix consists of two parts. 1) Teach LiveVariables that some unallocatable
registers might be liveouts so don't mark their last use as kill if they are.
2) ARM constantpool island pass shouldn't form cbz / cbnz if the conditional
branch does not kill CPSR.

rdar://10676853


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@148168 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 130 insertions, 6 deletions

diff --git a/lib/CodeGen/LiveVariables.cpp b/lib/CodeGen/LiveVariables.cpp
index 7477d91930..96d36c8e56 100644
--- a/lib/CodeGen/LiveVariables.cpp
+++ b/lib/CodeGen/LiveVariables.cpp
@@ -261,12 +261,11 @@ void LiveVariables::HandlePhysRegUse(unsigned Reg, MachineInstr *MI) {
           Processed.insert(*SS);
       }
     }
-  }
-  else if (LastDef && !PhysRegUse[Reg] &&
-           !LastDef->findRegisterDefOperand(Reg))
+  } else if (LastDef && !PhysRegUse[Reg] &&
+             !LastDef->findRegisterDefOperand(Reg))
     // Last def defines the super register, add an implicit def of reg.
-    LastDef->addOperand(MachineOperand::CreateReg(Reg,
-                                                 true/*IsDef*/, true/*IsImp*/));
+    LastDef->addOperand(MachineOperand::CreateReg(Reg, true/*IsDef*/,
+                                                  true/*IsImp*/));
 
   // Remember this use.
   PhysRegUse[Reg]  = MI;
@@ -607,10 +606,27 @@ bool LiveVariables::runOnMachineFunction(MachineFunction &mf) {
       }
     }
 
+    // MachineCSE may CSE instructions which write to non-allocatable physical
+    // registers across MBBs. Remember if any reserved register is liveout.
+    SmallSet<unsigned, 4> LiveOuts;
+    for (MachineBasicBlock::const_succ_iterator SI = MBB->succ_begin(),
+           SE = MBB->succ_end(); SI != SE; ++SI) {
+      MachineBasicBlock *SuccMBB = *SI;
+      if (SuccMBB->isLandingPad())
+        continue;
+      for (MachineBasicBlock::livein_iterator LI = SuccMBB->livein_begin(),
+             LE = SuccMBB->livein_end(); LI != LE; ++LI) {
+        unsigned LReg = *LI;
+        if (!TRI->isInAllocatableClass(LReg))
+          // Ignore other live-ins, e.g. those that are live into landing pads.
+          LiveOuts.insert(LReg);
+      }
+    }
+
     // Loop over PhysRegDef / PhysRegUse, killing any registers that are
     // available at the end of the basic block.
     for (unsigned i = 0; i != NumRegs; ++i)
-      if (PhysRegDef[i] || PhysRegUse[i])
+      if ((PhysRegDef[i] || PhysRegUse[i]) && !LiveOuts.count(i))
         HandlePhysRegDef(i, 0, Defs);
 
     std::fill(PhysRegDef,  PhysRegDef  + NumRegs, (MachineInstr*)0);
diff --git a/lib/Target/ARM/ARMConstantIslandPass.cpp b/lib/Target/ARM/ARMConstantIslandPass.cpp
index 26cf8622f3..f8936bc63f 100644
--- a/lib/Target/ARM/ARMConstantIslandPass.cpp
+++ b/lib/Target/ARM/ARMConstantIslandPass.cpp
@@ -1791,6 +1791,11 @@ bool ARMConstantIslands::OptimizeThumb2Branches() {
     if (Opcode != ARM::tBcc)
       continue;
 
+    // If the conditional branch doesn't kill CPSR, then CPSR can be liveout
+    // so this transformation is not safe.
+    if (!Br.MI->killsRegister(ARM::CPSR))
+      continue;
+
     NewOpc = 0;
     unsigned PredReg = 0;
     ARMCC::CondCodes Pred = llvm::getInstrPredicate(Br.MI, PredReg);
diff --git a/test/CodeGen/Thumb2/2012-01-13-CBNZBug.ll b/test/CodeGen/Thumb2/2012-01-13-CBNZBug.ll
new file mode 100644
index 0000000000..4acdd9e19e
--- /dev/null
+++ b/test/CodeGen/Thumb2/2012-01-13-CBNZBug.ll
@@ -0,0 +1,103 @@
+; RUN: llc < %s -mtriple=thumbv7-apple-ios -relocation-model=pic -disable-fp-elim -mcpu=cortex-a8 | FileCheck %s
+; rdar://10676853
+
+%struct.Dict_node_struct = type { i8*, %struct.Word_file_struct*, %struct.Exp_struct*, %struct.Dict_node_struct*, %struct.Dict_node_struct* }
+%struct.Word_file_struct = type { [60 x i8], i32, %struct.Word_file_struct* }
+%struct.Exp_struct = type { i8, i8, i8, i8, %union.anon }
+%union.anon = type { %struct.E_list_struct* }
+%struct.E_list_struct = type { %struct.E_list_struct*, %struct.Exp_struct* }
+
+@lookup_list = external hidden unnamed_addr global %struct.Dict_node_struct*, align 4
+
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
+
+define hidden fastcc void @rdictionary_lookup(%struct.Dict_node_struct* %dn, i8* nocapture %s) nounwind ssp {
+; CHECK: rdictionary_lookup:
+entry:
+  br label %tailrecurse
+
+tailrecurse:                                      ; preds = %if.then10, %entry
+  %dn.tr = phi %struct.Dict_node_struct* [ %dn, %entry ], [ %9, %if.then10 ]
+  %cmp = icmp eq %struct.Dict_node_struct* %dn.tr, null
+  br i1 %cmp, label %if.end11, label %if.end
+
+if.end:                                           ; preds = %tailrecurse
+  %string = getelementptr inbounds %struct.Dict_node_struct* %dn.tr, i32 0, i32 0
+  %0 = load i8** %string, align 4
+  br label %while.cond.i
+
+while.cond.i:                                     ; preds = %while.body.i, %if.end
+  %1 = phi i8* [ %s, %if.end ], [ %incdec.ptr.i, %while.body.i ]
+  %storemerge.i = phi i8* [ %0, %if.end ], [ %incdec.ptr6.i, %while.body.i ]
+  %2 = load i8* %1, align 1
+  %cmp.i = icmp eq i8 %2, 0
+  %.pre.i = load i8* %storemerge.i, align 1
+  br i1 %cmp.i, label %lor.lhs.false.i, label %land.end.i
+
+land.end.i:                                       ; preds = %while.cond.i
+  %cmp4.i = icmp eq i8 %2, %.pre.i
+  br i1 %cmp4.i, label %while.body.i, label %while.end.i
+
+while.body.i:                                     ; preds = %land.end.i
+  %incdec.ptr.i = getelementptr inbounds i8* %1, i32 1
+  %incdec.ptr6.i = getelementptr inbounds i8* %storemerge.i, i32 1
+  br label %while.cond.i
+
+while.end.i:                                      ; preds = %land.end.i
+  %cmp8.i = icmp eq i8 %2, 42
+  br i1 %cmp8.i, label %if.end3, label %lor.lhs.false.i
+
+lor.lhs.false.i:                                  ; preds = %while.end.i, %while.cond.i
+  %3 = phi i8 [ %2, %while.end.i ], [ 0, %while.cond.i ]
+  %cmp11.i = icmp eq i8 %.pre.i, 42
+  br i1 %cmp11.i, label %if.end3, label %dict_match.exit
+
+dict_match.exit:                                  ; preds = %lor.lhs.false.i
+  %cmp14.i = icmp eq i8 %3, 46
+  %conv16.i = sext i8 %3 to i32
+  %.conv16.i = select i1 %cmp14.i, i32 0, i32 %conv16.i
+  %cmp18.i = icmp eq i8 %.pre.i, 46
+  %conv22.i = sext i8 %.pre.i to i32
+  %cond24.i = select i1 %cmp18.i, i32 0, i32 %conv22.i
+  %sub.i = sub nsw i32 %.conv16.i, %cond24.i
+  %cmp1 = icmp sgt i32 %sub.i, -1
+  br i1 %cmp1, label %if.end3, label %if.then10
+
+if.end3:                                          ; preds = %dict_match.exit, %lor.lhs.false.i, %while.end.i
+; CHECK: %if.end3
+; CHECK: cmp
+; CHECK-NOT: cbnz
+  %storemerge1.i3 = phi i32 [ %sub.i, %dict_match.exit ], [ 0, %lor.lhs.false.i ], [ 0, %while.end.i ]
+  %right = getelementptr inbounds %struct.Dict_node_struct* %dn.tr, i32 0, i32 4
+  %4 = load %struct.Dict_node_struct** %right, align 4
+  tail call fastcc void @rdictionary_lookup(%struct.Dict_node_struct* %4, i8* %s)
+  %cmp4 = icmp eq i32 %storemerge1.i3, 0
+  br i1 %cmp4, label %if.then5, label %if.end8
+
+if.then5:                                         ; preds = %if.end3
+  %call6 = tail call fastcc i8* @xalloc(i32 20)
+  %5 = bitcast i8* %call6 to %struct.Dict_node_struct*
+  %6 = bitcast %struct.Dict_node_struct* %dn.tr to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i32(i8* %call6, i8* %6, i32 16, i32 4, i1 false)
+  %7 = load %struct.Dict_node_struct** @lookup_list, align 4
+  %right7 = getelementptr inbounds i8* %call6, i32 16
+  %8 = bitcast i8* %right7 to %struct.Dict_node_struct**
+  store %struct.Dict_node_struct* %7, %struct.Dict_node_struct** %8, align 4
+  store %struct.Dict_node_struct* %5, %struct.Dict_node_struct** @lookup_list, align 4
+  br label %if.then10
+
+if.end8:                                          ; preds = %if.end3
+  %cmp9 = icmp slt i32 %storemerge1.i3, 1
+  br i1 %cmp9, label %if.then10, label %if.end11
+
+if.then10:                                        ; preds = %if.end8, %if.then5, %dict_match.exit
+  %left = getelementptr inbounds %struct.Dict_node_struct* %dn.tr, i32 0, i32 3
+  %9 = load %struct.Dict_node_struct** %left, align 4
+  br label %tailrecurse
+
+if.end11:                                         ; preds = %if.end8, %tailrecurse
+  ret void
+}
+
+; Materializable
+declare hidden fastcc i8* @xalloc(i32) nounwind ssp