[SLP]Cleanup MultiNodeScalars when tree deleted.

Need to clear MultiNodeScalars map to avoid compiler crash when tree is
deleted.

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -1183,6 +1183,7 @@ class BoUpSLP {
   void deleteTree() {
     VectorizableTree.clear();
     ScalarToTreeEntry.clear();
+    MultiNodeScalars.clear();
     MustGather.clear();
     EntryToLastInstruction.clear();
     ExternalUses.clear();
@@ -2431,7 +2432,7 @@ class BoUpSLP {
       return false;
     });
     if (It != VL.end()) {
-      assert(TE->isSame(VL) && "Expedted same scalars.");
+      assert(TE->isSame(VL) && "Expected same scalars.");
       return TE;
     }
     return nullptr;

llvm/test/Transforms/SLPVectorizer/X86/reused-scalar-in-multi-node.ll

@@ -0,0 +1,52 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+%struct.UFP = type { i32, i32, i32, [4 x i32] }
+
+define void @test(ptr %u) {
+; CHECK-LABEL: define void @test(
+; CHECK-SAME: ptr [[U:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i32, ptr null, align 4
+; CHECK-NEXT:    [[IDX:%.*]] = getelementptr [[STRUCT_UFP:%.*]], ptr [[U]], i64 0, i32 3, i64 3
+; CHECK-NEXT:    [[IDX1:%.*]] = getelementptr [[STRUCT_UFP]], ptr [[U]], i64 0, i32 3, i64 2
+; CHECK-NEXT:    [[IDX2:%.*]] = load i32, ptr [[IDX]], align 4
+; CHECK-NEXT:    [[IDX3:%.*]] = load i32, ptr [[IDX1]], align 4
+; CHECK-NEXT:    br label [[WHILE:%.*]]
+; CHECK:       bb:
+; CHECK-NEXT:    store i32 [[OR_I_I:%.*]], ptr [[IDX]], align 4
+; CHECK-NEXT:    store i32 [[OR19_I_I:%.*]], ptr [[IDX1]], align 4
+; CHECK-NEXT:    ret void
+; CHECK:       while:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ [[TMP0]], [[ENTRY:%.*]] ], [ [[OR26_I_I:%.*]], [[WHILE]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = phi i32 [ [[IDX3]], [[ENTRY]] ], [ [[OR19_I_I]], [[WHILE]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = phi i32 [ [[IDX2]], [[ENTRY]] ], [ 0, [[WHILE]] ]
+; CHECK-NEXT:    [[OR_I_I]] = tail call i32 @llvm.fshl.i32(i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[OR19_I_I]] = tail call i32 @llvm.fshl.i32(i32 [[TMP1]], i32 [[TMP2]], i32 0)
+; CHECK-NEXT:    [[OR26_I_I]] = tail call i32 @llvm.fshl.i32(i32 0, i32 [[TMP1]], i32 0)
+; CHECK-NEXT:    br i1 false, label [[BB:%.*]], label [[WHILE]]
+;
+entry:
+  %0 = load i32, ptr null, align 4
+  %idx = getelementptr %struct.UFP, ptr %u, i64 0, i32 3, i64 3
+  %idx1 = getelementptr %struct.UFP, ptr %u, i64 0, i32 3, i64 2
+  %idx2 = load i32, ptr %idx, align 4
+  %idx3 = load i32, ptr %idx1, align 4
+  br label %while
+
+bb:
+  store i32 %or.i.i, ptr %idx, align 4
+  store i32 %or19.i.i, ptr %idx1, align 4
+  ret void
+
+while:
+  %1 = phi i32 [ %0, %entry ], [ %or26.i.i, %while ]
+  %2 = phi i32 [ %idx3, %entry ], [ %or19.i.i, %while ]
+  %3 = phi i32 [ %idx2, %entry ], [ 0, %while ]
+  %or.i.i = tail call i32 @llvm.fshl.i32(i32 %2, i32 0, i32 0)
+  %or19.i.i = tail call i32 @llvm.fshl.i32(i32 %1, i32 %2, i32 0)
+  %or26.i.i = tail call i32 @llvm.fshl.i32(i32 0, i32 %1, i32 0)
+  br i1 false, label %bb, label %while
+}
+
+declare i32 @llvm.fshl.i32(i32, i32, i32)