[SLP]Gather copyable node, if its parent is copyable, but this node is still used outside of the block only

If the current node is a copyable node and its parent is copyable too
and still current node is only used outside, better to cancel scheduling
for such node, because otherwise there might be wrong def-use chain
  built during vectorization.

Fixes llvm#166775

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -20996,6 +20996,15 @@ BoUpSLP::BlockScheduling::tryScheduleBundle(ArrayRef<Value *> VL, BoUpSLP *SLP,
         return false;
       }))
     return std::nullopt;
+  if (S.areInstructionsWithCopyableElements() && EI && EI.UserTE->hasState() &&
+      EI.UserTE->hasCopyableElements() &&
+      EI.UserTE->getMainOp()->getParent() == S.getMainOp()->getParent() &&
+      all_of(VL, [&](Value *V) {
+        if (S.isCopyableElement(V))
+          return true;
+        return isUsedOutsideBlock(V);
+      }))
+    return std::nullopt;
   bool HasCopyables = S.areInstructionsWithCopyableElements();
   if (((!HasCopyables && doesNotNeedToSchedule(VL)) ||
        all_of(VL, [&](Value *V) { return S.isNonSchedulable(V); }))) {

llvm/test/Transforms/SLPVectorizer/X86/copyable-child-node-used-outside.ll

@@ -0,0 +1,37 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes=slp-vectorizer -S -slp-threshold=-99999 -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
+
+define <4 x i32> @test() {
+; CHECK-LABEL: define <4 x i32> @test() {
+; CHECK-NEXT:  [[BB:.*:]]
+; CHECK-NEXT:    [[TRUNC:%.*]] = trunc i64 0 to i32
+; CHECK-NEXT:    br label %[[BB1:.*]]
+; CHECK:       [[BB1]]:
+; CHECK-NEXT:    [[OR:%.*]] = or i32 [[TRUNC]], 0
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <4 x i32> poison, i32 [[TRUNC]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x i32> [[TMP0]], i32 0, i32 1
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 0>
+; CHECK-NEXT:    [[TMP3:%.*]] = or <4 x i32> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[ZEXT:%.*]] = zext i32 [[OR]] to i64
+; CHECK-NEXT:    br label %[[BB3:.*]]
+; CHECK:       [[BB3]]:
+; CHECK-NEXT:    ret <4 x i32> [[TMP3]]
+;
+bb:
+  %trunc = trunc i64 0 to i32
+  br label %bb1
+
+bb1:
+  %or = or i32 %trunc, 0
+  %zext = zext i32 %or to i64
+  %and = and i32 0, 0
+  %or2 = or i32 %trunc, 0
+  br label %bb3
+
+bb3:
+  %0 = insertelement <4 x i32> zeroinitializer, i32 %trunc, i32 0
+  %1 = insertelement <4 x i32> %0, i32 %and, i32 1
+  %2 = insertelement <4 x i32> %1, i32 %or2, i32 2
+  %3 = insertelement <4 x i32> %2, i32 %or, i32 3
+  ret <4 x i32> %3
+}