[llvm] [CGP] [CodeGenPrepare] Folding `urem` with loop invariant value plus offset (PR #104724)

[via llvm-commits]

================
@@ -1982,17 +1982,36 @@ static bool foldFCmpToFPClassTest(CmpInst *Cmp, const TargetLowering &TLI,
   return true;
 }
 
-static bool isRemOfLoopIncrementWithLoopInvariant(Instruction *Rem,
-                                                  const LoopInfo *LI,
-                                                  Value *&RemAmtOut,
-                                                  PHINode *&LoopIncrPNOut) {
+static bool isRemOfLoopIncrementWithLoopInvariant(
+    Instruction *Rem, const LoopInfo *LI, Value *&RemAmtOut, Value *&AddInstOut,
+    Value *&AddOffsetOut, PHINode *&LoopIncrPNOut) {
   Value *Incr, *RemAmt;
   // NB: If RemAmt is a power of 2 it *should* have been transformed by now.
   if (!match(Rem, m_URem(m_Value(Incr), m_Value(RemAmt))))
     return false;
 
+  Value *AddInst, *AddOffset;
   // Find out loop increment PHI.
   auto *PN = dyn_cast<PHINode>(Incr);
+  if (PN != nullptr) {
+    AddInst = nullptr;
+    AddOffset = nullptr;
+  } else {
+    // Search through a NUW add on top of the loop increment.
+    Value *V0, *V1;
+    if (!match(Incr, m_NUWAdd(m_Value(V0), m_Value(V1))))
----------------
goldsteinn wrote:

We could, although Im not sure exactly what we would want in a PHI matcher? Would we support matching BBs? Variable number of arguments or "binary" phis only? Etc...
Either way, I think that its a conversation that is a bit beyond the scope of this particular use case and PR.

https://github.com/llvm/llvm-project/pull/104724