[SLP]Fix PR104637: do not create new nodes for fully overlapped non-schedulable nodes

If the scalars do not require scheduling and were already vectorized,
but in the different order, compiler still tries to create the new node.
It may cause the compiler crash for the gathered operands. Instead need
to consider such nodes as full overlap and just reshuffle vectorized
node.

Fixes #104637

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -7144,6 +7144,21 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
                        ReuseShuffleIndices);
         return;
       }
+      SmallPtrSet<const TreeEntry *, 4> Nodes;
+      Nodes.insert(getTreeEntry(S.OpValue));
+      for (const TreeEntry *E : MultiNodeScalars.lookup(S.OpValue))
+        Nodes.insert(E);
+      SmallPtrSet<Value *, 8> Values(VL.begin(), VL.end());
+      if (any_of(Nodes, [&](const TreeEntry *E) {
+            return all_of(E->Scalars,
+                          [&](Value *V) { return Values.contains(V); });
+          })) {
+        LLVM_DEBUG(dbgs() << "SLP: Gathering due to full overlap.\n");
+        if (TryToFindDuplicates(S))
+          newTreeEntry(VL, std::nullopt /*not vectorized*/, S, UserTreeIdx,
+                       ReuseShuffleIndices);
+        return;
+      }
     } else {
       // Record the reuse of the tree node.  FIXME, currently this is only used
       // to properly draw the graph rather than for the actual vectorization.

llvm/test/Transforms/SLPVectorizer/X86/full-non-schedulable-overlap.ll

@@ -0,0 +1,42 @@
+; 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 void @test(double %v) {
+; CHECK-LABEL: define void @test(
+; CHECK-SAME: double [[V:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x double> <double 0.000000e+00, double poison>, double [[V]], i32 1
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul <2 x double> zeroinitializer, [[TMP0]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> poison, <2 x i32> <i32 1, i32 0>
+; CHECK-NEXT:    br label %[[LOOP:.*]]
+; CHECK:       [[LOOP]]:
+; CHECK-NEXT:    [[T50_02:%.*]] = phi double [ 0.000000e+00, %[[ENTRY]] ], [ [[TMP8:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = phi <2 x double> [ zeroinitializer, %[[ENTRY]] ], [ [[TMP10:%.*]], %[[LOOP]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = fmul <2 x double> [[TMP3]], [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
+; CHECK-NEXT:    [[TMP7:%.*]] = fadd double [[TMP6]], [[TMP5]]
+; CHECK-NEXT:    [[TMP8]] = fadd double [[TMP7]], [[V]]
+; CHECK-NEXT:    [[TMP9:%.*]] = fmul <2 x double> zeroinitializer, [[TMP1]]
+; CHECK-NEXT:    [[TMP10]] = fadd <2 x double> zeroinitializer, [[TMP9]]
+; CHECK-NEXT:    br label %[[LOOP]]
+;
+entry:
+  %mul.3 = fmul double 0.000000e+00, %v
+  %mul.4 = fmul double 0.000000e+00, 0.000000e+00
+  br label %loop
+
+loop:
+  %t48.0 = phi double [ 0.000000e+00, %entry ], [ %5, %loop ]
+  %t50.02 = phi double [ 0.000000e+00, %entry ], [ %3, %loop ]
+  %t52.0 = phi double [ 0.000000e+00, %entry ], [ %7, %loop ]
+  %0 = fmul double %t52.0, %mul.3
+  %1 = fmul double %t48.0, %mul.4
+  %2 = fadd double %1, %0
+  %3 = fadd double %2, %v
+  %4 = fmul double 0.000000e+00, %mul.3
+  %5 = fadd double 0.000000e+00, %4
+  %6 = fmul double 0.000000e+00, %mul.4
+  %7 = fadd double 0.000000e+00, %6
+  br label %loop
+}