[llvm] [NaryReassociate] Teach NaryReassociate about UniformityAnalysis (PR #175167)

[Matt Arsenault via llvm-commits]

================
@@ -379,6 +392,33 @@ NaryReassociatePass::tryReassociateGEPAtIndex(GetElementPtrInst *GEP,
 
     Value *LHS = AO->getOperand(0), *RHS = AO->getOperand(1);
     // IndexToSplit = LHS + RHS.
+    // tryReassociateGEPAtIndex(GEP, I, LHS, RHS, ...) looks for a dominating
+    // GEP with LHS as index, then creates: NewGEP = existingGEP + RHS * scale.
+    // So the RHS becomes the "remaining" index calculation.
+    //
+    // For uniformity: prefer the remaining calculation to be uniform, as it
+    // can then stay in scalar registers.
+    //
+    // Default order tries LHS first (RHS as remainder). If LHS is uniform and
+    // RHS is divergent, we want to try RHS first so uniform LHS becomes the
+    // remainder. The case where RHS is uniform and LHS is divergent is already
+    // handled by the default order.
+    if (UI && UI->isUniform(LHS) && !UI->isUniform(RHS)) {
+      LLVM_DEBUG(
+          dbgs() << "NARY: Preferring uniform remainder for GEP index\n");
+      // LHS is uniform, prefer it as remainder - try RHS first
+      if (LHS != RHS) {
+        if (auto *NewGEP =
----------------
arsenm wrote:

No auto 

https://github.com/llvm/llvm-project/pull/175167