vendor/llvm/llvm-trunk-r257626

1 files changed, 11 insertions, 3 deletions

diff --git a/lib/Target/AMDGPU/SIFoldOperands.cpp b/lib/Target/AMDGPU/SIFoldOperands.cpp
index 02a39307e74e7..6230d1e28b74f 100644
--- a/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -334,12 +334,20 @@ bool SIFoldOperands::runOnMachineFunction(MachineFunction &MF) {
           !MRI.hasOneUse(MI.getOperand(0).getReg()))
         continue;
 
-      // FIXME: Fold operands with subregs.
       if (OpToFold.isReg() &&
-          (!TargetRegisterInfo::isVirtualRegister(OpToFold.getReg()) ||
-           OpToFold.getSubReg()))
+          !TargetRegisterInfo::isVirtualRegister(OpToFold.getReg()))
         continue;
 
+      // Prevent folding operands backwards in the function. For example,
+      // the COPY opcode must not be replaced by 1 in this example:
+      //
+      //    %vreg3<def> = COPY %VGPR0; VGPR_32:%vreg3
+      //    ...
+      //    %VGPR0<def> = V_MOV_B32_e32 1, %EXEC<imp-use>
+      MachineOperand &Dst = MI.getOperand(0);
+      if (Dst.isReg() &&
+          !TargetRegisterInfo::isVirtualRegister(Dst.getReg()))
+        continue;
 
       // We need mutate the operands of new mov instructions to add implicit
       // uses of EXEC, but adding them invalidates the use_iterator, so defer