[AMDGPU] Prevent hang in SIFoldOperands by caching Uses

foldOperands() for REG_SEQUENCE has recursion that can trigger infinite loop
as the method can modify use operand order which messes up the range-based
for loop. Cache the uses for processing beforehand so that iterators don't
get messed up.
Added repro mir testcase.

Author: choikwa

llvm/lib/Target/AMDGPU/SIFoldOperands.cpp

@@ -219,10 +219,8 @@ bool SIFoldOperands::canUseImmWithOpSel(FoldCandidate &Fold) const {
   default:
     return false;
   case AMDGPU::OPERAND_REG_IMM_V2FP16:
-  case AMDGPU::OPERAND_REG_IMM_V2BF16:
   case AMDGPU::OPERAND_REG_IMM_V2INT16:
   case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
-  case AMDGPU::OPERAND_REG_INLINE_C_V2BF16:
   case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
     break;
   }
@@ -774,21 +772,24 @@ void SIFoldOperands::foldOperand(
   if (UseMI->isRegSequence()) {
     Register RegSeqDstReg = UseMI->getOperand(0).getReg();
     unsigned RegSeqDstSubReg = UseMI->getOperand(UseOpIdx + 1).getImm();
-
-    for (auto &RSUse : make_early_inc_range(MRI->use_nodbg_operands(RegSeqDstReg))) {
-      MachineInstr *RSUseMI = RSUse.getParent();
+    
+    // Grab the use operands first
+    SmallVector<MachineOperand *, 4> UsesToProcess;
+    for (auto &Use : MRI->use_nodbg_operands(RegSeqDstReg))
+      UsesToProcess.push_back(&Use);
+    for (auto *RSUse : UsesToProcess) {
+      MachineInstr *RSUseMI = RSUse->getParent();
 
       if (tryToFoldACImm(UseMI->getOperand(0), RSUseMI,
-                         RSUseMI->getOperandNo(&RSUse), FoldList))
+                         RSUseMI->getOperandNo(RSUse), FoldList))
         continue;
 
-      if (RSUse.getSubReg() != RegSeqDstSubReg)
+      if (RSUse->getSubReg() != RegSeqDstSubReg)
         continue;
 
-      foldOperand(OpToFold, RSUseMI, RSUseMI->getOperandNo(&RSUse), FoldList,
+      foldOperand(OpToFold, RSUseMI, RSUseMI->getOperandNo(RSUse), FoldList,
                   CopiesToReplace);
     }
-
     return;
   }
 

llvm/test/CodeGen/AMDGPU/si-fold-reg-sequence.mir

@@ -0,0 +1,16 @@
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=si-fold-operands -verify-machineinstrs -o - %s
+
+---
+name:            fold_reg_sequence
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2
+    %33:sreg_32 = S_MOV_B32 0
+    %34:sreg_32 = S_MOV_B32 429
+    %35:sreg_64 = REG_SEQUENCE killed %34, %subreg.sub0, %33, %subreg.sub1
+    %49:vgpr_32 = V_MUL_HI_U32_e64 $vgpr2, %35.sub0, implicit $exec
+    %75:vgpr_32 = GLOBAL_LOAD_DWORD $vgpr0_vgpr1, 0, 0, implicit $exec :: (load (s32), addrspace 1)
+    %77:vgpr_32 = V_MUL_HI_U32_e64 %75, %35.sub0, implicit $exec
+    S_ENDPGM 0
+...
+