[LV] Return optz VPlan when trying to build (NFC)

[artagnon]

No description provided.

[llvmbot]

@llvm/pr-subscribers-vectorizers

@llvm/pr-subscribers-llvm-transforms

Author: Ramkumar Ramachandra (artagnon)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/158630.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/LoopVectorize.cpp (+7-13)
• (modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+6)

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 640a98c622f80..975b21a773c18 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -8229,20 +8229,8 @@ void LoopVectorizationPlanner::buildVPlansWithVPRecipes(ElementCount MinVF,
   for (ElementCount VF = MinVF; ElementCount::isKnownLT(VF, MaxVFTimes2);) {
     VFRange SubRange = {VF, MaxVFTimes2};
     if (auto Plan = tryToBuildVPlanWithVPRecipes(
-            std::unique_ptr<VPlan>(VPlan0->duplicate()), SubRange, &LVer)) {
-      bool HasScalarVF = Plan->hasScalarVFOnly();
-      // Now optimize the initial VPlan.
-      if (!HasScalarVF)
-        VPlanTransforms::runPass(VPlanTransforms::truncateToMinimalBitwidths,
-                                 *Plan, CM.getMinimalBitwidths());
-      VPlanTransforms::runPass(VPlanTransforms::optimize, *Plan);
-      // TODO: try to put it close to addActiveLaneMask().
-      if (CM.foldTailWithEVL() && !HasScalarVF)
-        VPlanTransforms::runPass(VPlanTransforms::addExplicitVectorLength,
-                                 *Plan, CM.getMaxSafeElements());
-      assert(verifyVPlanIsValid(*Plan) && "VPlan is invalid");
+            std::unique_ptr<VPlan>(VPlan0->duplicate()), SubRange, &LVer))
       VPlans.push_back(std::move(Plan));
-    }
     VF = SubRange.End;
   }
 }
@@ -8709,6 +8697,12 @@ VPlanPtr LoopVectorizationPlanner::tryToBuildVPlanWithVPRecipes(
                                        WithoutRuntimeCheck);
   }
   VPlanTransforms::optimizeInductionExitUsers(*Plan, IVEndValues, *PSE.getSE());
+  VPlanTransforms::runPass(VPlanTransforms::truncateToMinimalBitwidths, *Plan,
+                           CM.getMinimalBitwidths());
+  VPlanTransforms::runPass(VPlanTransforms::optimize, *Plan);
+  if (CM.foldTailWithEVL())
+    VPlanTransforms::runPass(VPlanTransforms::addExplicitVectorLength, *Plan,
+                             CM.getMaxSafeElements());
 
   assert(verifyVPlanIsValid(*Plan) && "VPlan is invalid");
   return Plan;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 503140213c116..59978f276db03 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2104,6 +2104,9 @@ static void licm(VPlan &Plan) {
 
 void VPlanTransforms::truncateToMinimalBitwidths(
     VPlan &Plan, const MapVector<Instruction *, uint64_t> &MinBWs) {
+  if (Plan.hasScalarVFOnly())
+    return;
+
   // Keep track of created truncates, so they can be re-used. Note that we
   // cannot use RAUW after creating a new truncate, as this would could make
   // other uses have different types for their operands, making them invalidly
@@ -2681,6 +2684,9 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
 ///
 void VPlanTransforms::addExplicitVectorLength(
     VPlan &Plan, const std::optional<unsigned> &MaxSafeElements) {
+  if (Plan.hasScalarVFOnly())
+    return;
+
   VPBasicBlock *Header = Plan.getVectorLoopRegion()->getEntryBasicBlock();
 
   auto *CanonicalIVPHI = Plan.getCanonicalIV();

[fhahn]

optimize is intentionally separate from tryToBuildVPlanWithVPRecipes, which ideally should only take care of constructing the initial widened VPlan, to be optimize later