[llvm] [LV] Optimize VPWidenIntOrFpInductionRecipe for known TC (PR #118828)

[Hari Limaye via llvm-commits]

================
@@ -661,11 +663,63 @@ static void recursivelyDeleteDeadRecipes(VPValue *V) {
   }
 }
 
-void VPlanTransforms::optimizeForVFAndUF(VPlan &Plan, ElementCount BestVF,
-                                         unsigned BestUF,
-                                         PredicatedScalarEvolution &PSE) {
-  assert(Plan.hasVF(BestVF) && "BestVF is not available in Plan");
-  assert(Plan.hasUF(BestUF) && "BestUF is not available in Plan");
+/// Optimize the width of vector induction variables based on \p TC, \p BestVF
+/// and \p BestUF.
+static bool optimizeVectorInductionWidthForTCAndVFUF(VPlan &Plan,
+                                                     ElementCount BestVF,
+                                                     unsigned BestUF) {
+  auto *TC = dyn_cast_if_present<ConstantInt>(
+      Plan.getTripCount()->getUnderlyingValue());
+  if (!TC || !BestVF.isFixed())
+    return false;
+
+  // Calculate the widest type required for known TC, VF and UF.
+  uint64_t TCVal = TC->getZExtValue();
+  uint64_t Width = BestVF.getKnownMinValue() * BestUF;
+  uint64_t MaxVal = alignTo(TCVal, Width) - 1;
+  unsigned MaxActiveBits = Log2_64_Ceil(MaxVal);
+  unsigned NewBitWidth = std::max<unsigned>(PowerOf2Ceil(MaxActiveBits), 8);
+  LLVMContext &Ctx = Plan.getCanonicalIV()->getScalarType()->getContext();
+  auto *NewIVTy = IntegerType::get(Ctx, NewBitWidth);
+
+  bool MadeChange = false;
+
+  VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
+  for (VPRecipeBase &Phi : HeaderVPBB->phis()) {
+    auto *WideIV = dyn_cast<VPWidenIntOrFpInductionRecipe>(&Phi);
+    if (!WideIV || !WideIV->isCanonical() ||
+        WideIV->hasMoreThanOneUniqueUser() ||
+        NewIVTy == WideIV->getScalarType())
+      continue;
+
+    // Currently only handle cases where the single user is a header-mask
+    // comparison with the backedge-taken-count.
+    VPValue *Bound;
+    using namespace VPlanPatternMatch;
+    if (!match(*WideIV->user_begin(),
+               m_Binary<Instruction::ICmp>(m_Specific(WideIV),
+                                           m_VPValue(Bound))) ||
+        Bound != Plan.getOrCreateBackedgeTakenCount())
----------------
hazzlim wrote:

Thanks, that's much neater  - Done.

https://github.com/llvm/llvm-project/pull/118828