[SLP]Do not remove reduced value, if it is a copyable

If the value is checked for the reduction and it is a copyable element
in a root node, it should not be deleted, since it may still be used
after vectorization.

Fixes #155512

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -20529,7 +20529,9 @@ Value *BoUpSLP::vectorizeTree(
           !(GatheredLoadsEntriesFirst.has_value() &&
             IE->Idx >= *GatheredLoadsEntriesFirst &&
             VectorizableTree.front()->isGather() &&
-            is_contained(VectorizableTree.front()->Scalars, I)))
+            is_contained(VectorizableTree.front()->Scalars, I)) &&
+          !(!VectorizableTree.front()->isGather() &&
+            VectorizableTree.front()->isCopyableElement(I)))
         continue;
       SmallVector<SelectInst *> LogicalOpSelects;
       I->replaceUsesWithIf(PoisonValue::get(I->getType()), [&](Use &U) {

llvm/test/Transforms/SLPVectorizer/RISCV/reduced-copyable-element.ll

@@ -0,0 +1,39 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux-gnu -mattr=+v < %s | FileCheck %s
+
+@n = external global [0 x i64]
+
+define i32 @main() {
+; CHECK-LABEL: define i32 @main(
+; CHECK-SAME: ) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[TMP0:%.*]] = call <2 x i64> @llvm.experimental.vp.strided.load.v2i64.p0.i64(ptr align 8 @n, i64 32, <2 x i1> splat (i1 true), i32 2)
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc <2 x i64> [[TMP0]] to <2 x i32>
+; CHECK-NEXT:    [[TMP2:%.*]] = call <2 x i32> @llvm.smin.v2i32(<2 x i32> [[TMP1]], <2 x i32> zeroinitializer)
+; CHECK-NEXT:    [[TMP3:%.*]] = sext <2 x i32> [[TMP2]] to <2 x i64>
+; CHECK-NEXT:    [[TMP4:%.*]] = call <2 x i64> @llvm.umin.v2i64(<2 x i64> [[TMP3]], <2 x i64> splat (i64 17179869184))
+; CHECK-NEXT:    [[TMP5:%.*]] = trunc <2 x i64> [[TMP4]] to <2 x i32>
+; CHECK-NEXT:    [[TMP6:%.*]] = add <2 x i32> [[TMP5]], <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @llvm.vector.reduce.or.v2i32(<2 x i32> [[TMP6]])
+; CHECK-NEXT:    ret i32 [[TMP7]]
+;
+entry:
+  %0 = load i64, ptr getelementptr (i8, ptr @n, i64 32), align 8
+  %conv13.i.1 = trunc i64 %0 to i32
+  %cond.i.1 = tail call i32 @llvm.smin.i32(i32 %conv13.i.1, i32 0)
+  %conv40.i.1 = sext i32 %cond.i.1 to i64
+  %cond47.i.1 = tail call i64 @llvm.umin.i64(i64 %conv40.i.1, i64 17179869184)
+  %1 = trunc i64 %cond47.i.1 to i32
+  %2 = add i32 %1, 1
+  %3 = load i64, ptr @n, align 8
+  %conv13.i.2 = trunc i64 %3 to i32
+  %cond.i.2 = tail call i32 @llvm.smin.i32(i32 %conv13.i.2, i32 0)
+  %conv40.i.2 = sext i32 %cond.i.2 to i64
+  %cond47.i.2 = tail call i64 @llvm.umin.i64(i64 %conv40.i.2, i64 17179869184)
+  %4 = trunc i64 %cond47.i.2 to i32
+  %5 = or i32 %2, %4
+  ret i32 %5
+}
+
+declare i32 @llvm.smin.i32(i32, i32)
+declare i64 @llvm.umin.i64(i64, i64)