[llvm] [AMDGPU] Delete redundant s_or_b32 (PR #165261)

[Matt Arsenault via llvm-commits]

================
@@ -10689,6 +10689,32 @@ bool SIInstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, Register SrcReg,
     if (!optimizeSCC(Def, &CmpInstr, RI))
       return false;
 
+    // If s_or_b32 result, sY, is unused (i.e. it is effectively a 64-bit
+    // s_cmp_lg of a register pair) and the inputs are the hi and lo-halves of a
+    // 64-bit foldableSelect then delete s_or_b32 in the sequence:
+    //    sX = s_cselect_b64 (non-zero imm), 0
+    //    sLo = copy sX.sub0
+    //    sHi = copy sX.sub1
+    //    sY = s_or_b32 sLo, sHi
+    if (Def->getOpcode() == AMDGPU::S_OR_B32 &&
+        MRI->use_nodbg_empty(Def->getOperand(0).getReg())) {
+      const MachineOperand &OrOpnd1 = Def->getOperand(1);
+      const MachineOperand &OrOpnd2 = Def->getOperand(2);
+      if (OrOpnd1.isReg() && OrOpnd2.isReg()) {
+        MachineInstr *Def1 = MRI->getUniqueVRegDef(OrOpnd1.getReg());
+        MachineInstr *Def2 = MRI->getUniqueVRegDef(OrOpnd2.getReg());
----------------
arsenm wrote:

Need to check if there failed, this will crash on undef operands. Probably should also use getVRegDef, this doesn't run on post-SSA 

https://github.com/llvm/llvm-project/pull/165261