[llvm] [GISel][AArch64][AMDGPU][RISCV] Canonicalize (sub X, C) -> (add X, -C) (PR #114309)

[Craig Topper via llvm-commits]

================
@@ -2041,6 +2041,28 @@ void CombinerHelper::applyCombineMulToShl(MachineInstr &MI,
   Observer.changedInstr(MI);
 }
 
+bool CombinerHelper::matchCombineSubToAdd(MachineInstr &MI,
+                                          BuildFnTy &MatchInfo) {
+  GSub &Sub = cast<GSub>(MI);
+
+  LLT Ty = MRI.getType(Sub.getReg(0));
+
+  if (!isLegalOrBeforeLegalizer({TargetOpcode::G_ADD, {Ty}}))
+    return false;
+
+  APInt Imm = getIConstantFromReg(Sub.getRHSReg(), MRI);
+
+  MatchInfo = [=, &MI](MachineIRBuilder &B) {
+    auto NegCst = B.buildConstant(Ty, -Imm);
----------------
topperc wrote:

I don't know what you mean. I checked G_CONSTANT in the .td file like you told me to.

https://github.com/llvm/llvm-project/pull/114309