[SLP]Fix insert point for copyable node with the last inst, used only outside the block

If the copyable entry has the last instruction, used only outside the
block, tha insert ion point for the vector code should be the last
instruction itself, not the following one. It prevents wrong def-use
sequences, which might be generated for the buildvector nodes.

Fixes #163404

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -17632,7 +17632,9 @@ void BoUpSLP::setInsertPointAfterBundle(const TreeEntry *E) {
   }
   if (IsPHI ||
       (!E->isGather() && E->State != TreeEntry::SplitVectorize &&
-       E->doesNotNeedToSchedule()) ||
+       (E->doesNotNeedToSchedule() ||
+        (E->hasCopyableElements() && !E->isCopyableElement(LastInst) &&
+         isUsedOutsideBlock(LastInst)))) ||
       (GatheredLoadsEntriesFirst.has_value() &&
        E->Idx >= *GatheredLoadsEntriesFirst && !E->isGather() &&
        E->getOpcode() == Instruction::Load)) {

llvm/test/Transforms/SLPVectorizer/X86/last-non-copyable-inst-used-outside-bb.ll

@@ -0,0 +1,89 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu -slp-threshold=-99999 < %s | FileCheck %s
+
+define void @test() {
+; CHECK-LABEL: define void @test() {
+; CHECK-NEXT:  [[BB:.*]]:
+; CHECK-NEXT:    br label %[[BB1:.*]]
+; CHECK:       [[BB1]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = phi <4 x i32> [ zeroinitializer, %[[BB]] ], [ [[TMP7:%.*]], %[[BB16:.*]] ], [ zeroinitializer, %[[BB1]] ]
+; CHECK-NEXT:    br i1 false, label %[[BB1]], label %[[BB5:.*]]
+; CHECK:       [[BB5]]:
+; CHECK-NEXT:    [[PHI8:%.*]] = phi double [ 0.000000e+00, %[[BB16]] ], [ 0.000000e+00, %[[BB1]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = phi <4 x i32> [ [[TMP8:%.*]], %[[BB16]] ], [ <i32 poison, i32 poison, i32 0, i32 0>, %[[BB1]] ]
+; CHECK-NEXT:    switch i32 0, label %[[BB21:.*]] [
+; CHECK-NEXT:      i32 4, label %[[BB21]]
+; CHECK-NEXT:      i32 1, label %[[BB21]]
+; CHECK-NEXT:      i32 0, label %[[BB9:.*]]
+; CHECK-NEXT:    ]
+; CHECK:       [[BB9]]:
+; CHECK-NEXT:    [[PHI13:%.*]] = phi double [ 0.000000e+00, %[[BB21]] ], [ 0.000000e+00, %[[BB5]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = phi <4 x i32> [ [[TMP1]], %[[BB21]] ], [ <i32 poison, i32 poison, i32 0, i32 0>, %[[BB5]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = phi <4 x i32> [ [[TMP9:%.*]], %[[BB21]] ], [ <i32 poison, i32 poison, i32 0, i32 0>, %[[BB5]] ]
+; CHECK-NEXT:    switch i32 0, label %[[BB15:.*]] [
+; CHECK-NEXT:      i32 1, label %[[BB14:.*]]
+; CHECK-NEXT:      i32 0, label %[[BB16]]
+; CHECK-NEXT:    ]
+; CHECK:       [[BB14]]:
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> <i32 poison, i32 poison, i32 0, i32 poison>, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
+; CHECK-NEXT:    br label %[[BB16]]
+; CHECK:       [[BB15]]:
+; CHECK-NEXT:    [[TMP5:%.*]] = add <4 x i32> <i32 poison, i32 poison, i32 0, i32 0>, [[TMP2]]
+; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> [[TMP5]], <4 x i32> <i32 poison, i32 poison, i32 2, i32 7>
+; CHECK-NEXT:    br label %[[BB16]]
+; CHECK:       [[BB16]]:
+; CHECK-NEXT:    [[PHI20:%.*]] = phi double [ 0.000000e+00, %[[BB15]] ], [ 0.000000e+00, %[[BB14]] ], [ 0.000000e+00, %[[BB9]] ]
+; CHECK-NEXT:    [[TMP7]] = phi <4 x i32> [ [[TMP5]], %[[BB15]] ], [ [[TMP4]], %[[BB14]] ], [ <i32 poison, i32 poison, i32 0, i32 0>, %[[BB9]] ]
+; CHECK-NEXT:    [[TMP8]] = phi <4 x i32> [ [[TMP6]], %[[BB15]] ], [ [[TMP3]], %[[BB14]] ], [ <i32 poison, i32 poison, i32 0, i32 0>, %[[BB9]] ]
+; CHECK-NEXT:    br i1 false, label %[[BB5]], label %[[BB1]]
+; CHECK:       [[BB21]]:
+; CHECK-NEXT:    [[TMP9]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> <i32 poison, i32 poison, i32 0, i32 poison>, <4 x i32> <i32 0, i32 1, i32 6, i32 3>
+; CHECK-NEXT:    br label %[[BB9]]
+;
+bb:
+  br label %bb1
+
+bb1:
+  %phi = phi i32 [ 0, %bb ], [ 0, %bb1 ], [ %phi17, %bb16 ]
+  %phi2 = phi i32 [ 0, %bb ], [ 0, %bb1 ], [ %phi18, %bb16 ]
+  %phi3 = phi i32 [ 0, %bb ], [ poison, %bb16 ], [ 0, %bb1 ]
+  %phi4 = phi i32 [ 0, %bb ], [ poison, %bb16 ], [ 0, %bb1 ]
+  br i1 false, label %bb1, label %bb5
+
+bb5:
+  %phi6 = phi i32 [ %phi17, %bb16 ], [ 0, %bb1 ]
+  %phi7 = phi i32 [ %phi19, %bb16 ], [ 0, %bb1 ]
+  %phi8 = phi double [ 0.000000e+00, %bb16 ], [ 0.000000e+00, %bb1 ]
+  switch i32 0, label %bb21 [
+  i32 4, label %bb21
+  i32 1, label %bb21
+  i32 0, label %bb9
+  ]
+
+bb9:
+  %phi10 = phi i32 [ %phi6, %bb21 ], [ 0, %bb5 ]
+  %phi11 = phi i32 [ %phi7, %bb21 ], [ 0, %bb5 ]
+  %phi12 = phi i32 [ 0, %bb21 ], [ 0, %bb5 ]
+  %phi13 = phi double [ 0.000000e+00, %bb21 ], [ 0.000000e+00, %bb5 ]
+  switch i32 0, label %bb15 [
+  i32 1, label %bb14
+  i32 0, label %bb16
+  ]
+
+bb14:
+  br label %bb16
+
+bb15:
+  %add = add i32 0, %phi10
+  br label %bb16
+
+bb16:
+  %phi17 = phi i32 [ %add, %bb15 ], [ %phi10, %bb14 ], [ 0, %bb9 ]
+  %phi18 = phi i32 [ %phi11, %bb15 ], [ 0, %bb14 ], [ 0, %bb9 ]
+  %phi19 = phi i32 [ %phi12, %bb15 ], [ %phi12, %bb14 ], [ 0, %bb9 ]
+  %phi20 = phi double [ 0.000000e+00, %bb15 ], [ 0.000000e+00, %bb14 ], [ 0.000000e+00, %bb9 ]
+  br i1 false, label %bb5, label %bb1
+
+bb21:
+  br label %bb9
+}