2 files changed, 19 insertions, 12 deletions

diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index f6c2719dd84..bea61493812 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -1777,9 +1777,9 @@ let hasSideEffects = 1 in {
 // replaced with exec mask operations.
 def SI_MASK_BRANCH : PseudoInstSI <
   (outs), (ins brtarget:$target, SReg_64:$dst)> {
-  let isBranch = 1;
+  let isBranch = 0;
   let isTerminator = 1;
-  let isBarrier = 1;
+  let isBarrier = 0;
   let SALU = 1;
 }
 
diff --git a/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp b/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
index 0f9d0917fb7..23043ea0d9d 100644
--- a/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
+++ b/llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp
@@ -80,7 +80,7 @@ private:
 
   bool shouldSkip(MachineBasicBlock *From, MachineBasicBlock *To);
 
-  void Skip(MachineInstr &From, MachineOperand &To);
+  MachineInstr *Skip(MachineInstr &From, MachineOperand &To);
   bool skipIfDead(MachineInstr &MI, MachineBasicBlock &NextBB);
 
   void If(MachineInstr &MI);
@@ -182,14 +182,15 @@ bool SILowerControlFlow::shouldSkip(MachineBasicBlock *From,
   return false;
 }
 
-void SILowerControlFlow::Skip(MachineInstr &From, MachineOperand &To) {
-
+MachineInstr *SILowerControlFlow::Skip(MachineInstr &From, MachineOperand &To) {
   if (!shouldSkip(*From.getParent()->succ_begin(), To.getMBB()))
-    return;
+    return nullptr;
 
-  DebugLoc DL = From.getDebugLoc();
-  BuildMI(*From.getParent(), &From, DL, TII->get(AMDGPU::S_CBRANCH_EXECZ))
+  const DebugLoc &DL = From.getDebugLoc();
+  MachineInstr *Skip =
+    BuildMI(*From.getParent(), &From, DL, TII->get(AMDGPU::S_CBRANCH_EXECZ))
     .addOperand(To);
+  return Skip;
 }
 
 bool SILowerControlFlow::skipIfDead(MachineInstr &MI, MachineBasicBlock &NextBB) {
@@ -242,10 +243,13 @@ void SILowerControlFlow::If(MachineInstr &MI) {
           .addReg(AMDGPU::EXEC)
           .addReg(Reg);
 
-  Skip(MI, MI.getOperand(2));
+  MachineInstr *SkipInst = Skip(MI, MI.getOperand(2));
+
+  // Insert before the new branch instruction.
+  MachineInstr *InsPt = SkipInst ? SkipInst : &MI;
 
   // Insert a pseudo terminator to help keep the verifier happy.
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::SI_MASK_BRANCH))
+  BuildMI(MBB, InsPt, DL, TII->get(AMDGPU::SI_MASK_BRANCH))
     .addOperand(MI.getOperand(2))
     .addReg(Reg);
 
@@ -275,10 +279,13 @@ void SILowerControlFlow::Else(MachineInstr &MI) {
           .addReg(AMDGPU::EXEC)
           .addReg(Dst);
 
-  Skip(MI, MI.getOperand(2));
+  MachineInstr *SkipInst = Skip(MI, MI.getOperand(2));
+
+  // Insert before the new branch instruction.
+  MachineInstr *InsPt = SkipInst ? SkipInst : &MI;
 
   // Insert a pseudo terminator to help keep the verifier happy.
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::SI_MASK_BRANCH))
+  BuildMI(MBB, InsPt, DL, TII->get(AMDGPU::SI_MASK_BRANCH))
     .addOperand(MI.getOperand(2))
     .addReg(Dst);