diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
index 35594ecbde16..00a9ea445bcd 100644
--- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -1596,6 +1596,10 @@ bool SIFoldOperands::tryFoldRegSequence(MachineInstr &MI) {
 
   LLVM_DEBUG(dbgs() << "Folded " << *RS << " into " << *UseMI);
 
+  // Erase the REG_SEQUENCE eagerly, unless we followed a chain of COPY users,
+  // in which case we can erase them all later in runOnMachineFunction.
+  if (MRI->use_nodbg_empty(MI.getOperand(0).getReg()))
+    MI.eraseFromParentAndMarkDBGValuesForRemoval();
   return true;
 }
 
@@ -1786,8 +1790,24 @@ bool SIFoldOperands::runOnMachineFunction(MachineFunction &MF) {
 
       // If we managed to fold all uses of this copy then we might as well
       // delete it now.
-      if (MRI->use_nodbg_empty(MI.getOperand(0).getReg()))
-        MI.eraseFromParentAndMarkDBGValuesForRemoval();
+      // The only reason we need to follow chains of copies here is that
+      // tryFoldRegSequence looks forward through copies before folding a
+      // REG_SEQUENCE into its eventual users.
+      auto *InstToErase = &MI;
+      while (MRI->use_nodbg_empty(InstToErase->getOperand(0).getReg())) {
+        auto &SrcOp = InstToErase->getOperand(1);
+        auto SrcReg = SrcOp.isReg() ? SrcOp.getReg() : Register();
+        InstToErase->eraseFromParentAndMarkDBGValuesForRemoval();
+        InstToErase = nullptr;
+        if (!SrcReg || SrcReg.isPhysical())
+          break;
+        InstToErase = MRI->getVRegDef(SrcReg);
+        if (!InstToErase || !TII->isFoldableCopy(*InstToErase))
+          break;
+      }
+      if (InstToErase && InstToErase->isRegSequence() &&
+          MRI->use_nodbg_empty(InstToErase->getOperand(0).getReg()))
+        InstToErase->eraseFromParentAndMarkDBGValuesForRemoval();
     }
   }
   return true;