Index: lib/Target/X86/X86CondBrFolding.cpp
===================================================================
--- lib/Target/X86/X86CondBrFolding.cpp
+++ lib/Target/X86/X86CondBrFolding.cpp
@@ -184,15 +184,25 @@
 
 // Fix up any PHI node in the successor of MBB.
 static void fixPHIsInSucc(MachineBasicBlock *MBB, MachineBasicBlock *OldMBB,
-                          MachineBasicBlock *NewMBB) {
+                          MachineBasicBlock *NewMBB, bool NewSuccs) {
   if (NewMBB == OldMBB)
     return;
   for (auto MI = MBB->instr_begin(), ME = MBB->instr_end();
        MI != ME && MI->isPHI(); ++MI)
-    for (unsigned i = 2, e = MI->getNumOperands() + 1; i != e; i += 2) {
-      MachineOperand &MO = MI->getOperand(i);
-      if (MO.getMBB() == OldMBB)
+
+    for (unsigned i = MI->getNumOperands(); i != 1; i -= 2) {
+      MachineOperand &MO = MI->getOperand(i - 1);
+      if (MO.getMBB() != OldMBB)
+        continue;
+      // We would like to change incoming edge for this operand from OldMBB to
+      // NewMBB. If there are already PHIs from MBB to NewMBB, we simply remove
+      // them. See PR39658.
+      if (NewSuccs)
         MO.setMBB(NewMBB);
+      else {
+        MI->RemoveOperand(i - 1);
+        MI->RemoveOperand(i - 2);
+      }
     }
 }
 
@@ -237,10 +247,21 @@
     MBBInfo->FBB = NewDest;
     BrMI = &*UncondBrI;
   }
-  fixPHIsInSucc(NewDest, OrigDest, MBB);
+
+  auto SI =
+      llvm::find_if(MBB->successors(), [&NewDest](const MachineBasicBlock *SI) {
+        return SI == NewDest;
+      });
+  bool AddNewSuccs = false;
+  BranchProbability Prob = MBPI->getEdgeProbability(MBB, OrigDest);
+  if (SI == MBB->succ_end()) {
+    MBB->addSuccessor(NewDest);
+    AddNewSuccs = true;
+  } else
+    Prob += MBPI->getEdgeProbability(MBB, NewDest);
+  fixPHIsInSucc(NewDest, OrigDest, MBB, AddNewSuccs);
+  setBranchProb(MBB, NewDest, Prob);
   BrMI->eraseFromParent();
-  MBB->addSuccessor(NewDest);
-  setBranchProb(MBB, NewDest, MBPI->getEdgeProbability(MBB, OrigDest));
   MBB->removeSuccessor(OrigDest);
 }
 
@@ -329,8 +350,17 @@
     BuildMI(*RootMBB, UncondBrI, RootMBB->findDebugLoc(UncondBrI),
             TII->get(X86::JMP_1))
         .addMBB(TargetMBB);
-    RootMBB->addSuccessor(TargetMBB);
-    fixPHIsInSucc(TargetMBB, &MBB, RootMBB);
+
+    auto SI = llvm::find_if(
+        RootMBB->successors(),
+        [&TargetMBB](const MachineBasicBlock *SI) { return SI == TargetMBB; });
+    bool AddNewSuccs = false;
+    if (SI == RootMBB->succ_end()) {
+      RootMBB->addSuccessor(TargetMBB);
+      AddNewSuccs = true;
+    }
+    fixPHIsInSucc(TargetMBB, &MBB, RootMBB, AddNewSuccs);
+
     RootMBB->erase(UncondBrI);
   } else {
     replaceBrDest(RootMBB, RootMBBInfo->TBB, TargetMBB);
@@ -578,5 +608,10 @@
       &getAnalysis<MachineBranchProbabilityInfo>();
 
   X86CondBrFolding CondBr(TII, MBPI, MF);
-  return CondBr.optimize();
+  bool Ret = CondBr.optimize();
+#ifdef EXPENSIVE_CHECKS
+  if (Ret)
+    MF.verify();
+#endif
+  return Ret;
 }
Index: test/CodeGen/X86/pr39658.ll
===================================================================
--- test/CodeGen/X86/pr39658.ll
+++ test/CodeGen/X86/pr39658.ll
@@ -0,0 +1,508 @@
+; RUN: llc -mtriple=x86_64-linux-gnu -mcpu=sandybridge %s -o - -verify-machineinstrs
+
+@global = external global i8 addrspace(1)*
+@global.1 = external global i8 addrspace(1)*
+@global.2 = external global i64
+
+declare void @llvm.experimental.deoptimize.isVoid(...)
+
+; Function Attrs: uwtable
+define void @zot(i8* %arg) #0 gc "statepoint-example" {
+bb:
+  %tmp = alloca i1, align 1
+  %tmp1 = getelementptr inbounds i8, i8* %arg, i64 8
+  %tmp2 = bitcast i8* %tmp1 to i32*
+  %tmp3 = load i32, i32* %tmp2, align 4
+  %tmp4 = getelementptr inbounds i8, i8* %arg, i64 16
+  %tmp5 = bitcast i8* %tmp4 to i32*
+  %tmp6 = load i32, i32* %tmp5, align 4
+  %tmp7 = getelementptr inbounds i8, i8* %arg, i64 32
+  %tmp8 = bitcast i8* %tmp7 to i32*
+  %tmp9 = load i32, i32* %tmp8, align 4
+  %tmp10 = getelementptr inbounds i8, i8* %arg, i64 40
+  %tmp11 = bitcast i8* %tmp10 to i32*
+  %tmp12 = load i32, i32* %tmp11, align 4
+  store volatile i1 true, i1* %tmp, align 1
+  %tmp13 = load volatile i1, i1* %tmp, align 1
+  br i1 %tmp13, label %bb14, label %bb223
+
+bb14:                                             ; preds = %bb
+  %tmp15 = mul i32 %tmp3, 24
+  %tmp16 = mul i32 %tmp3, 576
+  %tmp17 = mul i32 %tmp3, 13824
+  %tmp18 = add i32 %tmp9, 1
+  %tmp19 = icmp sgt i32 %tmp18, 192
+  %tmp20 = select i1 %tmp19, i32 %tmp18, i32 192
+  %tmp21 = sub i32 %tmp20, %tmp9
+  %tmp22 = xor i32 %tmp9, -1
+  %tmp23 = add i32 %tmp20, %tmp22
+  %tmp24 = and i32 %tmp21, 3
+  %tmp25 = icmp eq i32 %tmp24, 0
+  br i1 %tmp25, label %bb99, label %bb26
+
+bb26:                                             ; preds = %bb14
+  %tmp27 = load i8 addrspace(1)*, i8 addrspace(1)** @global.1, align 8
+  %tmp28 = icmp eq i8 addrspace(1)* %tmp27, null
+  %tmp29 = ptrtoint i8 addrspace(1)* %tmp27 to i64, !verifier_exception !0
+  %tmp30 = xor i64 %tmp29, -1
+  %tmp31 = inttoptr i64 %tmp30 to i8 addrspace(1)*, !verifier_exception !0
+  %tmp32 = select i1 %tmp28, i8 addrspace(1)* null, i8 addrspace(1)* %tmp31
+  %tmp33 = load i64, i64* @global.2, align 8
+  %tmp34 = ptrtoint i8 addrspace(1)* %tmp32 to i64, !verifier_exception !0
+  %tmp35 = and i64 %tmp33, %tmp34
+  %tmp36 = icmp eq i64 %tmp35, 0
+  br i1 %tmp36, label %bb39, label %bb37, !prof !1
+
+bb37:                                             ; preds = %bb26
+  %tmp38 = call i8 addrspace(1)* @baz(i8 addrspace(1)* %tmp32, i8 addrspace(1)** nonnull @global.1) #5
+  br label %bb39
+
+bb39:                                             ; preds = %bb37, %bb26
+  %tmp40 = phi i8 addrspace(1)* [ %tmp38, %bb37 ], [ %tmp32, %bb26 ]
+  %tmp41 = getelementptr inbounds i8, i8 addrspace(1)* %tmp40, i64 800
+  %tmp42 = bitcast i8 addrspace(1)* %tmp41 to i64 addrspace(1)*
+  br label %bb43
+
+bb43:                                             ; preds = %bb93, %bb39
+  %tmp44 = phi i64 addrspace(1)* [ %tmp42, %bb39 ], [ %tmp58, %bb93 ]
+  %tmp45 = phi i8 addrspace(1)* [ %tmp40, %bb39 ], [ %tmp59, %bb93 ]
+  %tmp46 = phi i32 [ %tmp6, %bb39 ], [ %tmp94, %bb93 ]
+  %tmp47 = phi i32 [ %tmp9, %bb39 ], [ %tmp96, %bb93 ]
+  %tmp48 = phi i32 [ %tmp12, %bb39 ], [ 1, %bb93 ]
+  %tmp49 = phi i32 [ %tmp24, %bb39 ], [ %tmp97, %bb93 ]
+  %tmp50 = load volatile i32, i32 addrspace(256)* inttoptr (i64 104 to i32 addrspace(256)*), align 8
+  %tmp51 = icmp eq i32 %tmp50, 0
+  br i1 %tmp51, label %bb57, label %bb52, !prof !1
+
+bb52:                                             ; preds = %bb43
+  %tmp53 = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 2882400000, i32 0, void ()* nonnull @wobble, i32 0, i32 0, i32 0, i32 19, i32 0, i32 1, i32 0, i32 28, i32 0, i32 6, i32 0, i32 7, i8* null, i32 3, i32 %tmp3, i32 3, i32 %tmp46, i32 7, i8* null, i32 3, i32 %tmp47, i32 3, i32 %tmp48, i8 addrspace(1)* %tmp45) #6
+  %tmp54 = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %tmp53, i32 26, i32 26) ; (%tmp45, %tmp45)
+  %tmp55 = getelementptr inbounds i8, i8 addrspace(1)* %tmp54, i64 800
+  %tmp56 = bitcast i8 addrspace(1)* %tmp55 to i64 addrspace(1)*
+  br label %bb57
+
+bb57:                                             ; preds = %bb52, %bb43
+  %tmp58 = phi i64 addrspace(1)* [ %tmp44, %bb43 ], [ %tmp56, %bb52 ]
+  %tmp59 = phi i8 addrspace(1)* [ %tmp45, %bb43 ], [ %tmp54, %bb52 ]
+  %tmp60 = icmp sgt i32 %tmp48, 1
+  br i1 %tmp60, label %bb76, label %bb61, !prof !2
+
+bb61:                                             ; preds = %bb57
+  %tmp62 = load i64, i64 addrspace(1)* %tmp58, align 8
+  %tmp63 = or i32 %tmp48, %tmp46
+  %tmp64 = sext i32 %tmp63 to i64
+  %tmp65 = add i64 %tmp62, %tmp64
+  store i64 %tmp65, i64 addrspace(1)* %tmp58, align 8
+  %tmp66 = icmp eq i32 %tmp48, 1
+  br i1 %tmp66, label %bb76, label %bb67, !prof !3
+
+bb67:                                             ; preds = %bb67, %bb61
+  %tmp68 = phi i64 [ %tmp73, %bb67 ], [ %tmp65, %bb61 ]
+  %tmp69 = phi i32 [ %tmp70, %bb67 ], [ %tmp48, %bb61 ]
+  %tmp70 = add nsw i32 %tmp69, 1
+  %tmp71 = or i32 %tmp70, %tmp46
+  %tmp72 = sext i32 %tmp71 to i64
+  %tmp73 = add i64 %tmp68, %tmp72
+  %tmp74 = icmp sgt i32 %tmp69, -1
+  br i1 %tmp74, label %bb75, label %bb67, !prof !2, !llvm.loop !4
+
+bb75:                                             ; preds = %bb67
+  store i64 %tmp73, i64 addrspace(1)* %tmp58, align 8
+  br label %bb76
+
+bb76:                                             ; preds = %bb75, %bb61, %bb57
+  %tmp77 = icmp sgt i32 %tmp47, 43
+  br i1 %tmp77, label %bb93, label %bb78, !prof !6
+
+bb78:                                             ; preds = %bb76
+  %tmp79 = icmp eq i32 %tmp47, 43
+  br i1 %tmp79, label %bb90, label %bb80, !prof !3
+
+bb80:                                             ; preds = %bb78
+  %tmp81 = icmp slt i32 %tmp47, 42
+  br i1 %tmp81, label %bb82, label %bb90, !prof !7
+
+bb82:                                             ; preds = %bb80
+  %tmp83 = icmp eq i32 %tmp47, 41
+  br i1 %tmp83, label %bb90, label %bb84, !prof !6
+
+bb84:                                             ; preds = %bb84, %bb82
+  %tmp85 = phi i32 [ %tmp87, %bb84 ], [ %tmp17, %bb82 ]
+  %tmp86 = phi i32 [ %tmp88, %bb84 ], [ 41, %bb82 ]
+  %tmp87 = mul i32 %tmp85, 24
+  %tmp88 = add nsw i32 %tmp86, -1
+  %tmp89 = icmp sgt i32 %tmp88, %tmp47
+  br i1 %tmp89, label %bb84, label %bb90, !prof !7, !llvm.loop !8
+
+bb90:                                             ; preds = %bb84, %bb82, %bb80, %bb78
+  %tmp91 = phi i32 [ %tmp15, %bb78 ], [ %tmp16, %bb80 ], [ %tmp17, %bb82 ], [ %tmp87, %bb84 ]
+  %tmp92 = sext i32 %tmp47 to i64
+  br label %bb93
+
+bb93:                                             ; preds = %bb90, %bb76
+  %tmp94 = phi i32 [ %tmp3, %bb76 ], [ %tmp91, %bb90 ]
+  %tmp95 = phi i64 [ 44, %bb76 ], [ %tmp92, %bb90 ]
+  %tmp96 = add i32 %tmp47, 1
+  %tmp97 = add nsw i32 %tmp49, -1
+  %tmp98 = icmp eq i32 %tmp97, 0
+  br i1 %tmp98, label %bb99, label %bb43, !llvm.loop !9
+
+bb99:                                             ; preds = %bb93, %bb14
+  %tmp100 = phi i64 [ undef, %bb14 ], [ %tmp95, %bb93 ]
+  %tmp101 = phi i32 [ %tmp6, %bb14 ], [ %tmp94, %bb93 ]
+  %tmp102 = phi i32 [ %tmp9, %bb14 ], [ %tmp96, %bb93 ]
+  %tmp103 = phi i32 [ %tmp12, %bb14 ], [ 1, %bb93 ]
+  %tmp104 = icmp ult i32 %tmp23, 3
+  br i1 %tmp104, label %bb166, label %bb105
+
+bb105:                                            ; preds = %bb99
+  %tmp106 = load i8 addrspace(1)*, i8 addrspace(1)** @global.1, align 8
+  %tmp107 = icmp eq i8 addrspace(1)* %tmp106, null
+  %tmp108 = ptrtoint i8 addrspace(1)* %tmp106 to i64, !verifier_exception !0
+  %tmp109 = xor i64 %tmp108, -1
+  %tmp110 = inttoptr i64 %tmp109 to i8 addrspace(1)*, !verifier_exception !0
+  %tmp111 = select i1 %tmp107, i8 addrspace(1)* null, i8 addrspace(1)* %tmp110
+  %tmp112 = load i64, i64* @global.2, align 8
+  %tmp113 = ptrtoint i8 addrspace(1)* %tmp111 to i64, !verifier_exception !0
+  %tmp114 = and i64 %tmp112, %tmp113
+  %tmp115 = icmp eq i64 %tmp114, 0
+  br i1 %tmp115, label %bb118, label %bb116, !prof !1
+
+bb116:                                            ; preds = %bb105
+  %tmp117 = call i8 addrspace(1)* @baz(i8 addrspace(1)* %tmp111, i8 addrspace(1)** nonnull @global.1) #5
+  br label %bb118
+
+bb118:                                            ; preds = %bb116, %bb105
+  %tmp119 = phi i8 addrspace(1)* [ %tmp117, %bb116 ], [ %tmp111, %bb105 ]
+  %tmp120 = getelementptr inbounds i8, i8 addrspace(1)* %tmp119, i64 800
+  %tmp121 = bitcast i8 addrspace(1)* %tmp120 to i64 addrspace(1)*
+  %tmp122 = or i32 %tmp15, 1
+  %tmp123 = sext i32 %tmp122 to i64
+  %tmp124 = or i32 %tmp3, 1
+  %tmp125 = sext i32 %tmp124 to i64
+  br label %bb126
+
+bb126:                                            ; preds = %bb302, %bb118
+  %tmp127 = phi i64 addrspace(1)* [ %tmp121, %bb118 ], [ %tmp140, %bb302 ]
+  %tmp128 = phi i8 addrspace(1)* [ %tmp119, %bb118 ], [ %tmp141, %bb302 ]
+  %tmp129 = phi i32 [ %tmp101, %bb118 ], [ %tmp303, %bb302 ]
+  %tmp130 = phi i32 [ %tmp102, %bb118 ], [ %tmp305, %bb302 ]
+  %tmp131 = phi i32 [ %tmp103, %bb118 ], [ 1, %bb302 ]
+  %tmp132 = load volatile i32, i32 addrspace(256)* inttoptr (i64 104 to i32 addrspace(256)*), align 8
+  %tmp133 = icmp eq i32 %tmp132, 0
+  br i1 %tmp133, label %bb139, label %bb134, !prof !1
+
+bb134:                                            ; preds = %bb126
+  %tmp135 = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 2882400000, i32 0, void ()* nonnull @wobble, i32 0, i32 0, i32 0, i32 19, i32 0, i32 1, i32 0, i32 28, i32 0, i32 6, i32 0, i32 7, i8* null, i32 3, i32 %tmp3, i32 3, i32 %tmp129, i32 7, i8* null, i32 3, i32 %tmp130, i32 3, i32 %tmp131, i8 addrspace(1)* %tmp128) #6
+  %tmp136 = call coldcc i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %tmp135, i32 26, i32 26) ; (%tmp128, %tmp128)
+  %tmp137 = getelementptr inbounds i8, i8 addrspace(1)* %tmp136, i64 800
+  %tmp138 = bitcast i8 addrspace(1)* %tmp137 to i64 addrspace(1)*
+  br label %bb139
+
+bb139:                                            ; preds = %bb134, %bb126
+  %tmp140 = phi i64 addrspace(1)* [ %tmp127, %bb126 ], [ %tmp138, %bb134 ]
+  %tmp141 = phi i8 addrspace(1)* [ %tmp128, %bb126 ], [ %tmp136, %bb134 ]
+  %tmp142 = icmp sgt i32 %tmp131, 1
+  br i1 %tmp142, label %bb150, label %bb143, !prof !2
+
+bb143:                                            ; preds = %bb139
+  %tmp144 = load i64, i64 addrspace(1)* %tmp140, align 8
+  %tmp145 = or i32 %tmp131, %tmp129
+  %tmp146 = sext i32 %tmp145 to i64
+  %tmp147 = add i64 %tmp144, %tmp146
+  store i64 %tmp147, i64 addrspace(1)* %tmp140, align 8
+  %tmp148 = icmp eq i32 %tmp131, 1
+  br i1 %tmp148, label %bb150, label %bb215, !prof !3
+
+bb149:                                            ; preds = %bb215
+  store i64 %tmp221, i64 addrspace(1)* %tmp140, align 8
+  br label %bb150
+
+bb150:                                            ; preds = %bb149, %bb143, %bb139
+  %tmp151 = icmp sgt i32 %tmp130, 43
+  br i1 %tmp151, label %bb158, label %bb152, !prof !6
+
+bb152:                                            ; preds = %bb150
+  %tmp153 = icmp eq i32 %tmp130, 43
+  br i1 %tmp153, label %bb158, label %bb154, !prof !3
+
+bb154:                                            ; preds = %bb152
+  %tmp155 = icmp slt i32 %tmp130, 42
+  br i1 %tmp155, label %bb156, label %bb158, !prof !7
+
+bb156:                                            ; preds = %bb154
+  %tmp157 = icmp eq i32 %tmp130, 41
+  br i1 %tmp157, label %bb158, label %bb209, !prof !6
+
+bb158:                                            ; preds = %bb209, %bb156, %bb154, %bb152, %bb150
+  %tmp159 = phi i32 [ %tmp3, %bb150 ], [ %tmp15, %bb152 ], [ %tmp16, %bb154 ], [ %tmp17, %bb156 ], [ %tmp212, %bb209 ]
+  %tmp160 = add i32 %tmp130, 1
+  %tmp161 = load i64, i64 addrspace(1)* %tmp140, align 8
+  %tmp162 = or i32 %tmp159, 1
+  %tmp163 = sext i32 %tmp162 to i64
+  %tmp164 = add i64 %tmp161, %tmp163
+  store i64 %tmp164, i64 addrspace(1)* %tmp140, align 8
+  %tmp165 = icmp sgt i32 %tmp160, 43
+  br i1 %tmp165, label %bb251, label %bb227, !prof !6
+
+bb166:                                            ; preds = %bb302, %bb99
+  %tmp167 = phi i64 [ %tmp100, %bb99 ], [ %tmp304, %bb302 ]
+  %tmp168 = load i64, i64* bitcast (i8 addrspace(1)** @global to i64*), align 8
+  %tmp169 = xor i64 %tmp168, -1
+  %tmp170 = inttoptr i64 %tmp169 to i8 addrspace(1)*, !verifier_exception !0
+  %tmp171 = load i64, i64* @global.2, align 8
+  %tmp172 = and i64 %tmp171, %tmp169
+  %tmp173 = icmp eq i64 %tmp172, 0
+  br i1 %tmp173, label %bb176, label %bb174, !prof !1
+
+bb174:                                            ; preds = %bb166
+  %tmp175 = call i8 addrspace(1)* @baz(i8 addrspace(1)* %tmp170, i8 addrspace(1)** nonnull @global) #5
+  br label %bb176
+
+bb176:                                            ; preds = %bb174, %bb166
+  %tmp177 = phi i8 addrspace(1)* [ %tmp175, %bb174 ], [ %tmp170, %bb166 ]
+  %tmp178 = getelementptr inbounds i8, i8 addrspace(1)* %tmp177, i64 792
+  %tmp179 = bitcast i8 addrspace(1)* %tmp178 to i8 addrspace(1)* addrspace(1)*
+  %tmp180 = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %tmp179, align 8
+  %tmp181 = icmp eq i8 addrspace(1)* %tmp180, null
+  %tmp182 = ptrtoint i8 addrspace(1)* %tmp180 to i64, !verifier_exception !0
+  %tmp183 = xor i64 %tmp182, -1
+  %tmp184 = inttoptr i64 %tmp183 to i8 addrspace(1)*, !verifier_exception !0
+  %tmp185 = select i1 %tmp181, i8 addrspace(1)* null, i8 addrspace(1)* %tmp184
+  %tmp186 = ptrtoint i8 addrspace(1)* %tmp185 to i64, !verifier_exception !0
+  %tmp187 = and i64 %tmp171, %tmp186
+  %tmp188 = icmp eq i64 %tmp187, 0
+  br i1 %tmp188, label %bb191, label %bb189, !prof !1
+
+bb189:                                            ; preds = %bb176
+  %tmp190 = call i8 addrspace(1)* @pluto(i8 addrspace(1)* %tmp185, i8 addrspace(1)* addrspace(1)* %tmp179) #5
+  br label %bb191
+
+bb191:                                            ; preds = %bb189, %bb176
+  %tmp192 = phi i8 addrspace(1)* [ %tmp190, %bb189 ], [ %tmp185, %bb176 ]
+  %tmp193 = icmp eq i8 addrspace(1)* %tmp192, null
+  br i1 %tmp193, label %bb225, label %bb194, !prof !10, !make.implicit !11
+
+bb194:                                            ; preds = %bb191
+  %tmp195 = load i64, i64* bitcast (i8 addrspace(1)** @global.1 to i64*), align 8
+  %tmp196 = xor i64 %tmp195, -1
+  %tmp197 = inttoptr i64 %tmp196 to i8 addrspace(1)*, !verifier_exception !0
+  %tmp198 = and i64 %tmp171, %tmp196
+  %tmp199 = icmp eq i64 %tmp198, 0
+  br i1 %tmp199, label %bb202, label %bb200, !prof !1
+
+bb200:                                            ; preds = %bb194
+  %tmp201 = call i8 addrspace(1)* @baz(i8 addrspace(1)* %tmp197, i8 addrspace(1)** nonnull @global.1) #5
+  br label %bb202
+
+bb202:                                            ; preds = %bb200, %bb194
+  %tmp203 = phi i8 addrspace(1)* [ %tmp201, %bb200 ], [ %tmp197, %bb194 ]
+  %tmp204 = getelementptr inbounds i8, i8 addrspace(1)* %tmp203, i64 800
+  %tmp205 = bitcast i8 addrspace(1)* %tmp204 to i64 addrspace(1)*
+  %tmp206 = load i64, i64 addrspace(1)* %tmp205, align 8
+  %tmp207 = add i64 %tmp206, %tmp167
+  %tmp208 = call token (i64, i32, void (i8 addrspace(1)*, i64)*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidp1i8i64f(i64 2, i32 5, void (i8 addrspace(1)*, i64)* nonnull @blam, i32 2, i32 0, i8 addrspace(1)* nonnull %tmp192, i64 %tmp207, i32 0, i32 19, i32 0, i32 2, i32 0, i32 94, i32 0, i32 6, i32 0, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i8 addrspace(1)* nonnull %tmp192) #7
+  ret void
+
+bb209:                                            ; preds = %bb209, %bb156
+  %tmp210 = phi i32 [ %tmp212, %bb209 ], [ %tmp17, %bb156 ]
+  %tmp211 = phi i32 [ %tmp213, %bb209 ], [ 41, %bb156 ]
+  %tmp212 = mul i32 %tmp210, 24
+  %tmp213 = add nsw i32 %tmp211, -1
+  %tmp214 = icmp sgt i32 %tmp213, %tmp130
+  br i1 %tmp214, label %bb209, label %bb158, !prof !7, !llvm.loop !8
+
+bb215:                                            ; preds = %bb215, %bb143
+  %tmp216 = phi i64 [ %tmp221, %bb215 ], [ %tmp147, %bb143 ]
+  %tmp217 = phi i32 [ %tmp218, %bb215 ], [ %tmp131, %bb143 ]
+  %tmp218 = add nsw i32 %tmp217, 1
+  %tmp219 = or i32 %tmp218, %tmp129
+  %tmp220 = sext i32 %tmp219 to i64
+  %tmp221 = add i64 %tmp216, %tmp220
+  %tmp222 = icmp sgt i32 %tmp217, -1
+  br i1 %tmp222, label %bb149, label %bb215, !prof !2, !llvm.loop !4
+
+bb223:                                            ; preds = %bb
+  %tmp224 = call token (i64, i32, void (i32)*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidi32f(i64 2882400000, i32 0, void (i32)* nonnull @widget, i32 1, i32 0, i32 0, i32 0, i32 19, i32 0, i32 1, i32 0, i32 28, i32 0, i32 6, i32 0, i32 7, i8* null, i32 3, i32 %tmp3, i32 3, i32 %tmp6, i32 7, i8* null, i32 3, i32 %tmp9, i32 3, i32 %tmp12)
+  unreachable
+
+bb225:                                            ; preds = %bb191
+  %tmp226 = call token (i64, i32, void (i32)*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidi32f(i64 2882400000, i32 0, void (i32)* nonnull @widget, i32 1, i32 0, i32 8, i32 0, i32 25, i32 0, i32 1, i32 0, i32 94, i32 3, i32 6, i32 0, i32 0, i8 addrspace(1)* null, i32 7, i64 undef, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null, i32 7, i8* null)
+  unreachable
+
+bb227:                                            ; preds = %bb158
+  %tmp228 = icmp eq i32 %tmp160, 43
+  br i1 %tmp228, label %bb229, label %bb232, !prof !3
+
+bb229:                                            ; preds = %bb227
+  %tmp230 = add i64 %tmp164, %tmp123
+  %tmp231 = add i64 %tmp230, %tmp125
+  store i64 %tmp231, i64 addrspace(1)* %tmp140, align 8
+  br label %bb302
+
+bb232:                                            ; preds = %bb227
+  %tmp233 = icmp slt i32 %tmp160, 42
+  br i1 %tmp233, label %bb234, label %bb242, !prof !7
+
+bb234:                                            ; preds = %bb232
+  %tmp235 = icmp eq i32 %tmp160, 41
+  br i1 %tmp235, label %bb242, label %bb236, !prof !6
+
+bb236:                                            ; preds = %bb236, %bb234
+  %tmp237 = phi i32 [ %tmp239, %bb236 ], [ %tmp17, %bb234 ]
+  %tmp238 = phi i32 [ %tmp240, %bb236 ], [ 41, %bb234 ]
+  %tmp239 = mul i32 %tmp237, 24
+  %tmp240 = add nsw i32 %tmp238, -1
+  %tmp241 = icmp sgt i32 %tmp240, %tmp160
+  br i1 %tmp241, label %bb236, label %bb248, !prof !7, !llvm.loop !8
+
+bb242:                                            ; preds = %bb234, %bb232
+  %tmp243 = phi i32 [ %tmp17, %bb234 ], [ %tmp16, %bb232 ]
+  %tmp244 = add nuw nsw i32 %tmp130, 2
+  %tmp245 = or i32 %tmp243, 1
+  %tmp246 = sext i32 %tmp245 to i64
+  %tmp247 = add i64 %tmp164, %tmp246
+  store i64 %tmp247, i64 addrspace(1)* %tmp140, align 8
+  br label %bb256
+
+bb248:                                            ; preds = %bb236
+  %tmp249 = or i32 %tmp239, 1
+  %tmp250 = sext i32 %tmp249 to i64
+  br label %bb251
+
+bb251:                                            ; preds = %bb248, %bb158
+  %tmp252 = phi i64 [ %tmp250, %bb248 ], [ %tmp125, %bb158 ]
+  %tmp253 = add i32 %tmp130, 2
+  %tmp254 = add i64 %tmp164, %tmp252
+  store i64 %tmp254, i64 addrspace(1)* %tmp140, align 8
+  %tmp255 = icmp sgt i32 %tmp253, 43
+  br i1 %tmp255, label %bb279, label %bb256, !prof !6
+
+bb256:                                            ; preds = %bb251, %bb242
+  %tmp257 = phi i64 [ %tmp247, %bb242 ], [ %tmp254, %bb251 ]
+  %tmp258 = phi i32 [ %tmp244, %bb242 ], [ %tmp253, %bb251 ]
+  %tmp259 = icmp eq i32 %tmp258, 43
+  br i1 %tmp259, label %bb279, label %bb260, !prof !3
+
+bb260:                                            ; preds = %bb256
+  %tmp261 = icmp slt i32 %tmp258, 42
+  br i1 %tmp261, label %bb262, label %bb270, !prof !7
+
+bb262:                                            ; preds = %bb260
+  %tmp263 = icmp eq i32 %tmp258, 41
+  br i1 %tmp263, label %bb270, label %bb264, !prof !6
+
+bb264:                                            ; preds = %bb264, %bb262
+  %tmp265 = phi i32 [ %tmp267, %bb264 ], [ %tmp17, %bb262 ]
+  %tmp266 = phi i32 [ %tmp268, %bb264 ], [ 41, %bb262 ]
+  %tmp267 = mul i32 %tmp265, 24
+  %tmp268 = add nsw i32 %tmp266, -1
+  %tmp269 = icmp sgt i32 %tmp268, %tmp258
+  br i1 %tmp269, label %bb264, label %bb276, !prof !7, !llvm.loop !8
+
+bb270:                                            ; preds = %bb262, %bb260
+  %tmp271 = phi i32 [ %tmp17, %bb262 ], [ %tmp16, %bb260 ]
+  %tmp272 = add i32 %tmp130, 3
+  %tmp273 = or i32 %tmp271, 1
+  %tmp274 = sext i32 %tmp273 to i64
+  %tmp275 = add i64 %tmp257, %tmp274
+  store i64 %tmp275, i64 addrspace(1)* %tmp140, align 8
+  br label %bb285
+
+bb276:                                            ; preds = %bb264
+  %tmp277 = or i32 %tmp267, 1
+  %tmp278 = sext i32 %tmp277 to i64
+  br label %bb279
+
+bb279:                                            ; preds = %bb276, %bb256, %bb251
+  %tmp280 = phi i64 [ %tmp278, %bb276 ], [ %tmp123, %bb256 ], [ %tmp125, %bb251 ]
+  %tmp281 = phi i64 [ %tmp257, %bb276 ], [ %tmp257, %bb256 ], [ %tmp254, %bb251 ]
+  %tmp282 = add i32 %tmp130, 3
+  %tmp283 = add i64 %tmp281, %tmp280
+  store i64 %tmp283, i64 addrspace(1)* %tmp140, align 8
+  %tmp284 = icmp sgt i32 %tmp282, 43
+  br i1 %tmp284, label %bb302, label %bb285, !prof !6
+
+bb285:                                            ; preds = %bb279, %bb270
+  %tmp286 = phi i32 [ %tmp272, %bb270 ], [ %tmp282, %bb279 ]
+  %tmp287 = icmp eq i32 %tmp286, 43
+  br i1 %tmp287, label %bb298, label %bb288, !prof !3
+
+bb288:                                            ; preds = %bb285
+  %tmp289 = icmp slt i32 %tmp286, 42
+  br i1 %tmp289, label %bb290, label %bb298, !prof !7
+
+bb290:                                            ; preds = %bb288
+  %tmp291 = icmp eq i32 %tmp286, 41
+  br i1 %tmp291, label %bb298, label %bb292, !prof !6
+
+bb292:                                            ; preds = %bb292, %bb290
+  %tmp293 = phi i32 [ %tmp295, %bb292 ], [ %tmp17, %bb290 ]
+  %tmp294 = phi i32 [ %tmp296, %bb292 ], [ 41, %bb290 ]
+  %tmp295 = mul i32 %tmp293, 24
+  %tmp296 = add nsw i32 %tmp294, -1
+  %tmp297 = icmp sgt i32 %tmp296, %tmp286
+  br i1 %tmp297, label %bb292, label %bb298, !prof !7, !llvm.loop !8
+
+bb298:                                            ; preds = %bb292, %bb290, %bb288, %bb285
+  %tmp299 = phi i32 [ 43, %bb285 ], [ 42, %bb288 ], [ 41, %bb290 ], [ %tmp286, %bb292 ]
+  %tmp300 = phi i32 [ %tmp15, %bb285 ], [ %tmp16, %bb288 ], [ %tmp17, %bb290 ], [ %tmp295, %bb292 ]
+  %tmp301 = sext i32 %tmp299 to i64
+  br label %bb302
+
+bb302:                                            ; preds = %bb298, %bb279, %bb229
+  %tmp303 = phi i32 [ %tmp3, %bb279 ], [ %tmp300, %bb298 ], [ %tmp3, %bb229 ]
+  %tmp304 = phi i64 [ 44, %bb279 ], [ %tmp301, %bb298 ], [ 44, %bb229 ]
+  %tmp305 = add i32 %tmp130, 4
+  %tmp306 = icmp sgt i32 %tmp305, 191
+  br i1 %tmp306, label %bb166, label %bb126, !prof !12
+}
+
+; Function Attrs: nounwind
+declare void @wobble() #1
+
+; Function Attrs: noinline
+declare void @blam(i8 addrspace(1)*, i64) #2
+
+declare void @wibble(...)
+
+declare token @llvm.experimental.gc.statepoint.p0f_isVoidf(i64, i32, void ()*, i32, i32, ...)
+
+; Function Attrs: nounwind readonly
+declare i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token, i32, i32) #3
+
+declare token @llvm.experimental.gc.statepoint.p0f_isVoidp1i8i64f(i64, i32, void (i8 addrspace(1)*, i64)*, i32, i32, ...)
+
+declare void @widget(i32)
+
+declare token @llvm.experimental.gc.statepoint.p0f_isVoidi32f(i64, i32, void (i32)*, i32, i32, ...)
+
+; Function Attrs: argmemonly nounwind
+declare i8 addrspace(1)* @baz(i8 addrspace(1)*, i8 addrspace(1)**) #4
+
+; Function Attrs: argmemonly nounwind
+declare i8 addrspace(1)* @pluto(i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)*) #4
+
+attributes #0 = { uwtable "disable-tail-calls"="true" "gc-add-backedge-safepoints"="true" "gc-add-call-safepoints"="true" "gc-add-entry-safepoints"="true" "no-entry-safepoint-poll" "patchable-function"="prologue-short-redirect" "target-cpu"="broadwell" "target-features"="+sse2,+cx16,+sahf,-tbm,-avx512ifma,-sha,-gfni,-fma4,-vpclmulqdq,+prfchw,+bmi2,-cldemote,+fsgsbase,-ptwrite,-xsavec,+popcnt,+aes,-avx512bitalg,-movdiri,-xsaves,-avx512er,-avx512vnni,-avx512vpopcntdq,-pconfig,-clwb,-avx512f,-clzero,-pku,+mmx,-lwp,-rdpid,-xop,+rdseed,-waitpkg,-movdir64b,-sse4a,-avx512bw,-clflushopt,+xsave,-avx512vbmi2,+64bit,-avx512vl,+invpcid,-avx512cd,+avx,-vaes,+rtm,+fma,+bmi,+rdrnd,-mwaitx,+sse4.1,+sse4.2,+avx2,-wbnoinvd,+sse,+lzcnt,+pclmul,-prefetchwt1,+f16c,+ssse3,-sgx,-shstk,+cmov,-avx512vbmi,+movbe,+xsaveopt,-avx512dq,+adx,-avx512pf,+sse3" }
+attributes #1 = { nounwind "consumes-vmstate" "target-cpu"="broadwell" "target-features"="+sse2,+cx16,+sahf,-tbm,-avx512ifma,-sha,-gfni,-fma4,-vpclmulqdq,+prfchw,+bmi2,-cldemote,+fsgsbase,-ptwrite,-xsavec,+popcnt,+aes,-avx512bitalg,-movdiri,-xsaves,-avx512er,-avx512vnni,-avx512vpopcntdq,-pconfig,-clwb,-avx512f,-clzero,-pku,+mmx,-lwp,-rdpid,-xop,+rdseed,-waitpkg,-movdir64b,-sse4a,-avx512bw,-clflushopt,+xsave,-avx512vbmi2,+64bit,-avx512vl,+invpcid,-avx512cd,+avx,-vaes,+rtm,+fma,+bmi,+rdrnd,-mwaitx,+sse4.1,+sse4.2,+avx2,-wbnoinvd,+sse,+lzcnt,+pclmul,-prefetchwt1,+f16c,+ssse3,-sgx,-shstk,+cmov,-avx512vbmi,+movbe,+xsaveopt,-avx512dq,+adx,-avx512pf,+sse3" }
+attributes #2 = { noinline "bytecode-size"="26" "guaranteed-to-poll" }
+attributes #3 = { nounwind readonly }
+attributes #4 = { argmemonly nounwind "gc-leaf-function" "target-cpu"="broadwell" "target-features"="+sse2,+cx16,+sahf,-tbm,-avx512ifma,-sha,-gfni,-fma4,-vpclmulqdq,+prfchw,+bmi2,-cldemote,+fsgsbase,-ptwrite,-xsavec,+popcnt,+aes,-avx512bitalg,-movdiri,-xsaves,-avx512er,-avx512vnni,-avx512vpopcntdq,-pconfig,-clwb,-avx512f,-clzero,-pku,+mmx,-lwp,-rdpid,-xop,+rdseed,-waitpkg,-movdir64b,-sse4a,-avx512bw,-clflushopt,+xsave,-avx512vbmi2,+64bit,-avx512vl,+invpcid,-avx512cd,+avx,-vaes,+rtm,+fma,+bmi,+rdrnd,-mwaitx,+sse4.1,+sse4.2,+avx2,-wbnoinvd,+sse,+lzcnt,+pclmul,-prefetchwt1,+f16c,+ssse3,-sgx,-shstk,+cmov,-avx512vbmi,+movbe,+xsaveopt,-avx512dq,+adx,-avx512pf,+sse3" }
+attributes #5 = { cold nounwind "gc-leaf-function" }
+attributes #6 = { nounwind "inline-remark"="unavailable definition" }
+attributes #7 = { "callsite-frequency"="1" "inline-remark"="unavailable definition" }
+
+!0 = !{i1 true}
+!1 = !{!"branch_weights", i32 1048575, i32 1}
+!2 = !{!"branch_weights", i32 1, i32 1}
+!3 = !{!"branch_weights", i32 1, i32 0}
+!4 = distinct !{!4, !5}
+!5 = !{!"llvm.loop.unroll.disable"}
+!6 = !{!"branch_weights", i32 1, i32 3}
+!7 = !{!"branch_weights", i32 3, i32 1}
+!8 = distinct !{!8, !5}
+!9 = distinct !{!9, !5}
+!10 = !{!"branch_weights", i32 1, i32 983040}
+!11 = !{}
+!12 = !{!"branch_weights", i32 1, i32 176}
+