[SLP]Fix the minbitwidth analysis for slternate opcodes

If the laternate operation is more stricter than the main operation, we
cannot rely on the analysis of the main operation. In such case, better
to avoid doing the analysis at all, since it may affect the overall
result and lead to incorrect optimization

Fixes #165878

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -22134,6 +22134,27 @@ bool BoUpSLP::collectValuesToDemote(
         {VectorizableTree[E.CombinedEntriesWithIndices.front().first].get(),
          VectorizableTree[E.CombinedEntriesWithIndices.back().first].get()});
 
+  if (E.isAltShuffle()) {
+    // Combining these opcodes may lead to incorrect analysis, skip for now.
+    auto IsDangerousOpcode = [](unsigned Opcode) {
+      switch (Opcode) {
+      case Instruction::Shl:
+      case Instruction::AShr:
+      case Instruction::LShr:
+      case Instruction::UDiv:
+      case Instruction::SDiv:
+      case Instruction::URem:
+      case Instruction::SRem:
+        return true;
+      default:
+        break;
+      }
+      return false;
+    };
+    if (IsDangerousOpcode(E.getAltOpcode()))
+      return FinalAnalysis();
+  }
+
   switch (E.getOpcode()) {
 
   // We can always demote truncations and extensions. Since truncations can

llvm/test/Transforms/SLPVectorizer/X86/alternate-opcode-strict-bitwidth-than-main.ll

@@ -6,14 +6,12 @@ define float @test(i8 %0) {
 ; CHECK-SAME: i8 [[TMP0:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i8> <i8 poison, i8 0>, i8 [[TMP0]], i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = sext <2 x i8> [[TMP1]] to <2 x i16>
-; CHECK-NEXT:    [[TMP3:%.*]] = mul <2 x i16> [[TMP2]], <i16 2, i16 27>
-; CHECK-NEXT:    [[TMP4:%.*]] = lshr <2 x i16> [[TMP2]], <i16 2, i16 27>
-; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i16> [[TMP3]], <2 x i16> [[TMP4]], <2 x i32> <i32 0, i32 3>
-; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i16> [[TMP5]], i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = sext i16 [[TMP9]] to i32
-; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i16> [[TMP5]], i32 1
-; CHECK-NEXT:    [[TMP7:%.*]] = zext i16 [[TMP10]] to i32
+; CHECK-NEXT:    [[TMP2:%.*]] = sext <2 x i8> [[TMP1]] to <2 x i32>
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <2 x i32> [[TMP2]], <i32 2, i32 27>
+; CHECK-NEXT:    [[TMP4:%.*]] = lshr <2 x i32> [[TMP2]], <i32 2, i32 27>
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> [[TMP4]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i32> [[TMP5]], i32 0
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i32> [[TMP5]], i32 1
 ; CHECK-NEXT:    [[TMP8:%.*]] = or i32 [[TMP6]], [[TMP7]]
 ; CHECK-NEXT:    switch i32 [[TMP8]], label %[[EXIT:.*]] [
 ; CHECK-NEXT:      i32 0, label %[[EXIT]]