[SLP]Do not include copyable data to the same user twice

If the copyable schedule data is created and the user is used several
times in the user node, no need to count same data for the same user
several times, need to include it only ones.

Fixes llvm#153754

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -5336,6 +5336,7 @@ class BoUpSLP {
         ArrayRef<Value *> Op = EI.UserTE->getOperand(EI.EdgeIdx);
         const auto *It = find(Op, I);
         assert(It != Op.end() && "Lane not set");
+        SmallPtrSet<Instruction *, 4> Visited;
         do {
           int Lane = std::distance(Op.begin(), It);
           assert(Lane >= 0 && "Lane not set");
@@ -5345,6 +5346,10 @@ class BoUpSLP {
           assert(Lane < static_cast<int>(EI.UserTE->Scalars.size()) &&
                  "Couldn't find extract lane");
           auto *In = cast<Instruction>(EI.UserTE->Scalars[Lane]);
+          if (!Visited.insert(In).second) {
+            It = find(make_range(std::next(It), Op.end()), I);
+            continue;
+          }
           ScheduleCopyableDataMapByInstUser
               .try_emplace(std::make_pair(std::make_pair(In, EI.EdgeIdx), I))
               .first->getSecond()
@@ -20927,6 +20932,7 @@ BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP,
     }
     ScheduledBundlesList.pop_back();
     SmallVector<ScheduleData *> ControlDependentMembers;
+    SmallPtrSet<Instruction *, 4> Visited;
     for (Value *V : VL) {
       if (S.isNonSchedulable(V))
         continue;
@@ -20944,6 +20950,7 @@ BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP,
           ArrayRef<Value *> Op = EI.UserTE->getOperand(EI.EdgeIdx);
           const auto *It = find(Op, I);
           assert(It != Op.end() && "Lane not set");
+          SmallPtrSet<Instruction *, 4> Visited;
           do {
             int Lane = std::distance(Op.begin(), It);
             assert(Lane >= 0 && "Lane not set");
@@ -20953,6 +20960,10 @@ BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP,
             assert(Lane < static_cast<int>(EI.UserTE->Scalars.size()) &&
                    "Couldn't find extract lane");
             auto *In = cast<Instruction>(EI.UserTE->Scalars[Lane]);
+            if (!Visited.insert(In).second) {
+              It = find(make_range(std::next(It), Op.end()), I);
+              break;
+            }
             ScheduleCopyableDataMapByInstUser
                 [std::make_pair(std::make_pair(In, EI.EdgeIdx), I)]
                     .pop_back();

llvm/test/Transforms/SLPVectorizer/X86/schedule-same-user-with-copyable.ll

@@ -0,0 +1,64 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+define i32 @test(ptr %o, i32 %b.021.i) {
+; CHECK-LABEL: define i32 @test(
+; CHECK-SAME: ptr [[O:%.*]], i32 [[B_021_I:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[O1:%.*]] = alloca [3 x i32], align 4
+; CHECK-NEXT:    br label %[[WHILE_BODY:.*]]
+; CHECK:       [[WHILE_BODY]]:
+; CHECK-NEXT:    [[SUB623:%.*]] = phi i32 [ [[SUB6:%.*]], %[[N_EXIT:.*]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    [[ADD21:%.*]] = phi i32 [ [[ADD:%.*]], %[[N_EXIT]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    [[ADD419:%.*]] = phi i32 [ [[ADD4:%.*]], %[[N_EXIT]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT:    [[ADD18:%.*]] = phi i32 [ [[ADD]], %[[N_EXIT]] ], [ 1, %[[ENTRY]] ]
+; CHECK-NEXT:    store i32 [[ADD419]], ptr [[O1]], align 4
+; CHECK-NEXT:    store i32 [[ADD18]], ptr [[O]], align 4
+; CHECK-NEXT:    br label %[[FOR_BODY4_I:.*]]
+; CHECK:       [[FOR_COND1_I:.*]]:
+; CHECK-NEXT:    ret i32 0
+; CHECK:       [[FOR_BODY4_I]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr [[O1]], align 4
+; CHECK-NEXT:    [[TOBOOL_NOT_I:%.*]] = icmp eq i32 [[TMP0]], 0
+; CHECK-NEXT:    br i1 [[TOBOOL_NOT_I]], label %[[FOR_COND1_I]], label %[[N_EXIT]]
+; CHECK:       [[N_EXIT]]:
+; CHECK-NEXT:    [[SUB:%.*]] = or i32 [[B_021_I]], [[ADD21]]
+; CHECK-NEXT:    [[ADD]] = or i32 [[SUB]], 1
+; CHECK-NEXT:    [[ADD2:%.*]] = or i32 [[B_021_I]], 1
+; CHECK-NEXT:    [[ADD3:%.*]] = add i32 [[ADD2]], [[SUB623]]
+; CHECK-NEXT:    [[ADD4]] = or i32 [[ADD3]], 1
+; CHECK-NEXT:    [[MUL:%.*]] = shl i32 [[B_021_I]], 1
+; CHECK-NEXT:    [[SUB6]] = or i32 [[MUL]], 1
+; CHECK-NEXT:    br label %[[WHILE_BODY]]
+;
+entry:
+  %o1 = alloca [3 x i32], align 4
+  br label %while.body
+
+while.body:                                       ; preds = %n.exit, %entry
+  %sub623 = phi i32 [ %sub6, %n.exit ], [ 0, %entry ]
+  %add21 = phi i32 [ %add, %n.exit ], [ 0, %entry ]
+  %add419 = phi i32 [ %add4, %n.exit ], [ 0, %entry ]
+  %add18 = phi i32 [ %add, %n.exit ], [ 1, %entry ]
+  store i32 %add419, ptr %o1, align 4
+  store i32 %add18, ptr %o, align 4
+  br label %for.body4.i
+
+for.cond1.i:                                      ; preds = %for.body4.i
+  ret i32 0
+
+for.body4.i:                                      ; preds = %while.body
+  %0 = load i32, ptr %o1, align 4
+  %tobool.not.i = icmp eq i32 %0, 0
+  br i1 %tobool.not.i, label %for.cond1.i, label %n.exit
+
+n.exit:                                           ; preds = %for.body4.i
+  %sub = or i32 %b.021.i, %add21
+  %add = or i32 %sub, 1
+  %add2 = or i32 %b.021.i, 1
+  %add3 = add i32 %add2, %sub623
+  %add4 = or i32 %add3, 1
+  %mul = shl i32 %b.021.i, 1
+  %sub6 = or i32 %mul, 1
+  br label %while.body
+}