[VPlan] Remove trivial dead VPPhi cycles.

fhahn · fhahn · commit 1f17bb133f4f · 2025-08-11T21:29:49.000+01:00
Update removeDeadRecipes to remove trivial dead VPPhi cycles.

Should effectively be NFC end-to-end.
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7347,8 +7347,7 @@ DenseMap<const SCEV *, Value *> LoopVectorizationPlanner::executePlan(
   BasicBlock *EntryBB =
       cast<VPIRBasicBlock>(BestVPlan.getEntry())->getIRBasicBlock();
   State.CFG.PrevBB = ILV.createVectorizedLoopSkeleton();
-  if (VectorizingEpilogue)
-    VPlanTransforms::removeDeadRecipes(BestVPlan);
+  VPlanTransforms::removeDeadRecipes(BestVPlan);
 
   assert(verifyVPlanIsValid(BestVPlan, true /*VerifyLate*/) &&
          "final VPlan is invalid");
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -553,8 +553,19 @@ void VPlanTransforms::removeDeadRecipes(VPlan &Plan) {
     // The recipes in the block are processed in reverse order, to catch chains
     // of dead recipes.
     for (VPRecipeBase &R : make_early_inc_range(reverse(*VPBB))) {
-      if (isDeadRecipe(R))
+      if (isDeadRecipe(R)) {
         R.eraseFromParent();
+        continue;
+      }
+
+      // Check if R is a dead VPPhi <-> update cycle and remove it.
+      auto *PhiR = dyn_cast<VPPhi>(&R);
+      if (!PhiR || PhiR->getNumOperands() != 2 || PhiR->getNumUsers() != 1 ||
+          *PhiR->user_begin() != PhiR->getOperand(1)->getDefiningRecipe() ||
+          PhiR->getOperand(1)->getNumUsers() != 1)
+        continue;
+      PhiR->replaceAllUsesWith(PhiR->getOperand(0));
+      PhiR->eraseFromParent();
     }
   }
 }
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/force-target-instruction-cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/force-target-instruction-cost.ll
@@ -11,12 +11,10 @@ define double @test_reduction_costs() {
 ; CHECK:       [[VECTOR_PH]]:
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; CHECK:       [[VECTOR_BODY]]:
-; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi double [ 0.000000e+00, %[[VECTOR_PH]] ], [ [[TMP0:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi double [ 0.000000e+00, %[[VECTOR_PH]] ], [ [[TMP1:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP0]] = call double @llvm.vector.reduce.fadd.v2f64(double [[VEC_PHI]], <2 x double> splat (double 3.000000e+00))
 ; CHECK-NEXT:    [[TMP1]] = call double @llvm.vector.reduce.fadd.v2f64(double [[VEC_PHI1]], <2 x double> splat (double 9.000000e+00))
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[EXIT:.*]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll b/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll
@@ -1101,10 +1101,8 @@ define i64 @cost_loop_invariant_recipes(i1 %x, i64 %y) {
 ; CHECK-NEXT:    [[TMP2:%.*]] = shl <2 x i64> [[BROADCAST_SPLAT2]], [[TMP1]]
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
-; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ splat (i64 1), [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP3]] = mul <2 x i64> [[TMP2]], [[VEC_PHI]]
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    br i1 true, label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vector.reduce.mul.v2i64(<2 x i64> [[TMP3]])
@@ -1156,13 +1154,11 @@ define i32 @narrowed_reduction(ptr %a, i1 %cmp) #0 {
 ; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i32> [[BROADCAST_SPLATINSERT]], <16 x i32> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
-; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH1]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP1:%.*]] = and <16 x i32> [[VEC_PHI1]], splat (i32 1)
 ; CHECK-NEXT:    [[TMP3:%.*]] = or <16 x i32> [[TMP1]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP5:%.*]] = trunc <16 x i32> [[TMP3]] to <16 x i1>
 ; CHECK-NEXT:    [[TMP7]] = zext <16 x i1> [[TMP5]] to <16 x i32>
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 16
 ; CHECK-NEXT:    br i1 true, label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
 ; CHECK:       middle.block:
 ; CHECK-NEXT:    [[TMP20:%.*]] = call i1 @llvm.vector.reduce.or.v16i1(<16 x i1> [[TMP5]])
diff --git a/llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll b/llvm/test/Transforms/LoopVectorize/X86/induction-costs.ll
@@ -401,13 +401,11 @@ define i16 @iv_and_step_trunc() {
 ; CHECK:       vector.ph:
 ; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
 ; CHECK:       vector.body:
-; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[VEC_IND1:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP0:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 1)
 ; CHECK-NEXT:    [[TMP1:%.*]] = trunc <2 x i64> [[TMP0]] to <2 x i16>
 ; CHECK-NEXT:    [[TMP2:%.*]] = mul <2 x i16> [[VEC_IND1]], [[TMP1]]
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
 ; CHECK-NEXT:    [[VEC_IND_NEXT2]] = add <2 x i16> [[VEC_IND1]], splat (i16 2)
 ; CHECK-NEXT:    br i1 true, label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/outer_loop_test1_no_explicit_vect_width.ll b/llvm/test/Transforms/LoopVectorize/X86/outer_loop_test1_no_explicit_vect_width.ll
@@ -67,8 +67,7 @@ define void @foo(i32 %n) {
 ; AVX-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT]], <8 x i32> poison, <8 x i32> zeroinitializer
 ; AVX-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; AVX:       [[VECTOR_BODY]]:
-; AVX-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_LATCH:.*]] ]
-; AVX-NEXT:    [[VEC_IND:%.*]] = phi <8 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_LATCH]] ]
+; AVX-NEXT:    [[VEC_IND:%.*]] = phi <8 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_LATCH:.*]] ]
 ; AVX-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [8 x i32], ptr @arr2, i64 0, <8 x i64> [[VEC_IND]]
 ; AVX-NEXT:    [[TMP1:%.*]] = trunc <8 x i64> [[VEC_IND]] to <8 x i32>
 ; AVX-NEXT:    call void @llvm.masked.scatter.v8i32.v8p0(<8 x i32> [[TMP1]], <8 x ptr> [[TMP0]], i32 4, <8 x i1> splat (i1 true))
@@ -84,7 +83,6 @@ define void @foo(i32 %n) {
 ; AVX-NEXT:    [[TMP6:%.*]] = extractelement <8 x i1> [[TMP5]], i32 0
 ; AVX-NEXT:    br i1 [[TMP6]], label %[[VECTOR_LATCH]], label %[[FOR_BODY31]]
 ; AVX:       [[VECTOR_LATCH]]:
-; AVX-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
 ; AVX-NEXT:    [[VEC_IND_NEXT]] = add <8 x i64> [[VEC_IND]], splat (i64 8)
 ; AVX-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
 ; AVX:       [[MIDDLE_BLOCK]]:
