[SLP]Fix PR89438: Check for same vectorized node in MinBWs, not user.

Need to check if the buildvector node has perfect diamond match in the
graph and the matched node is resized.

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -13139,9 +13139,30 @@ Value *BoUpSLP::vectorizeTree(
       assert(Vec->getType()->isIntOrIntVectorTy() &&
              PrevVec->getType()->isIntOrIntVectorTy() &&
              "Expected integer vector types only.");
-      assert(MinBWs.contains(TE->UserTreeIndices.front().UserTE) &&
-             "Expected user in MinBWs.");
-      bool IsSigned = MinBWs.lookup(TE->UserTreeIndices.front().UserTE).second;
+      std::optional<std::pair<unsigned long, bool>> Res;
+      if (const TreeEntry *BaseTE = getTreeEntry(TE->Scalars.front())) {
+        SmallVector<const TreeEntry *> BaseTEs;
+        if (BaseTE->isSame(TE->Scalars))
+          BaseTEs.push_back(BaseTE);
+        auto It = MultiNodeScalars.find(TE->Scalars.front());
+        if (It != MultiNodeScalars.end()) {
+          for (const TreeEntry *MNTE : It->getSecond())
+            if (MNTE->isSame(TE->Scalars))
+              BaseTEs.push_back(MNTE);
+        }
+        const auto *BaseIt = find_if(BaseTEs, [&](const TreeEntry *BaseTE) {
+          return MinBWs.contains(BaseTE);
+        });
+        if (BaseIt != BaseTEs.end())
+          Res = MinBWs.lookup(*BaseIt);
+      }
+      if (!Res) {
+        assert(MinBWs.contains(TE->UserTreeIndices.front().UserTE) &&
+               "Expected user in MinBWs.");
+        Res = MinBWs.lookup(TE->UserTreeIndices.front().UserTE);
+      }
+      assert(Res && "Expected user node or perfect diamond match in MinBWs.");
+      bool IsSigned = Res->second;
       Vec = Builder.CreateIntCast(Vec, PrevVec->getType(), IsSigned);
     }
     PrevVec->replaceAllUsesWith(Vec);

llvm/test/Transforms/SLPVectorizer/X86/gather-node-same-reduced.ll

@@ -0,0 +1,84 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux < %s | FileCheck %s
+
+define i64 @test(ptr %p) {
+; CHECK-LABEL: define i64 @test(
+; CHECK-SAME: ptr [[P:%.*]]) {
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[P]], i64 12
+; CHECK-NEXT:    [[TMP2:%.*]] = xor <4 x i32> zeroinitializer, zeroinitializer
+; CHECK-NEXT:    [[TMP3:%.*]] = xor <4 x i32> [[TMP2]], zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = xor <4 x i32> [[TMP3]], zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = xor <4 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = xor <4 x i32> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = xor <4 x i32> [[TMP6]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = xor <4 x i32> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = xor <4 x i32> [[TMP8]], zeroinitializer
+; CHECK-NEXT:    [[TMP10:%.*]] = xor <4 x i32> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = xor <4 x i32> [[TMP10]], zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = trunc <4 x i32> [[TMP11]] to <4 x i8>
+; CHECK-NEXT:    store <4 x i8> [[TMP12]], ptr [[TMP1]], align 1
+; CHECK-NEXT:    ret i64 0
+;
+  %1 = getelementptr i8, ptr %p, i64 13
+  %2 = getelementptr i8, ptr %p, i64 14
+  %3 = getelementptr i8, ptr %p, i64 15
+  %4 = getelementptr i8, ptr %p, i64 12
+  %5 = zext i8 0 to i32
+  %6 = and i32 %5, 0
+  %.not866 = icmp eq i32 %6, 0
+  %7 = select i1 %.not866, i32 0, i32 0
+  %8 = xor i32 0, %7
+  %9 = zext i8 0 to i32
+  %10 = and i32 %9, 0
+  %.not871 = icmp eq i32 %10, 0
+  %11 = select i1 %.not871, i32 0, i32 0
+  %12 = xor i32 0, %11
+  %13 = xor i32 %9, 0
+  %14 = xor i32 %13, 0
+  %15 = xor i32 %14, 0
+  %16 = xor i32 %15, 0
+  %17 = xor i32 %16, 0
+  %18 = xor i32 %17, %12
+  %19 = xor i32 %18, 0
+  %20 = xor i32 %19, 0
+  %21 = xor i32 %20, 0
+  %22 = xor i32 %21, 0
+  %23 = trunc i32 %22 to i8
+  store i8 %23, ptr %4, align 1
+  %24 = xor i32 %9, 0
+  %25 = xor i32 %24, 0
+  %26 = xor i32 %25, 0
+  %27 = xor i32 %26, 0
+  %28 = xor i32 %27, 0
+  %29 = xor i32 %28, %8
+  %30 = xor i32 %29, 0
+  %31 = xor i32 %30, 0
+  %32 = xor i32 %31, 0
+  %33 = xor i32 %32, 0
+  %34 = trunc i32 %33 to i8
+  store i8 %34, ptr %1, align 1
+  %35 = xor i32 0, %5
+  %36 = xor i32 %35, 0
+  %37 = xor i32 %36, 0
+  %38 = xor i32 %37, 0
+  %39 = xor i32 %38, 0
+  %40 = xor i32 %39, %8
+  %41 = xor i32 %40, 0
+  %42 = xor i32 %41, 0
+  %43 = xor i32 %42, 0
+  %44 = xor i32 %43, 0
+  %45 = trunc i32 %44 to i8
+  store i8 %45, ptr %2, align 1
+  %46 = xor i32 %35, 0
+  %47 = xor i32 %46, 0
+  %48 = xor i32 %47, 0
+  %49 = xor i32 %48, 0
+  %50 = xor i32 %49, %8
+  %51 = xor i32 %50, 0
+  %52 = xor i32 %51, 0
+  %53 = xor i32 %52, 0
+  %54 = xor i32 %53, 0
+  %55 = trunc i32 %54 to i8
+  store i8 %55, ptr %3, align 1
+  ret i64 0
+}