[SLP][REVEC] Make computeMinimumValueSizes and collectValuesToDemote support vector instructions. (#103005)

Author: HanKuanChen

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -15564,7 +15564,8 @@ bool BoUpSLP::collectValuesToDemote(
   if (all_of(E.Scalars, IsaPred<Constant>))
     return true;
 
-  unsigned OrigBitWidth = DL->getTypeSizeInBits(E.Scalars.front()->getType());
+  unsigned OrigBitWidth =
+      DL->getTypeSizeInBits(E.Scalars.front()->getType()->getScalarType());
   if (OrigBitWidth == BitWidth) {
     MaxDepthLevel = 1;
     return true;
@@ -15995,7 +15996,8 @@ void BoUpSLP::computeMinimumValueSizes() {
     }
 
     unsigned VF = E.getVectorFactor();
-    auto *TreeRootIT = dyn_cast<IntegerType>(E.Scalars.front()->getType());
+    auto *TreeRootIT =
+        dyn_cast<IntegerType>(E.Scalars.front()->getType()->getScalarType());
     if (!TreeRootIT || !Opcode)
       return 0u;
 
@@ -16145,7 +16147,8 @@ void BoUpSLP::computeMinimumValueSizes() {
 
     for (unsigned Idx : RootDemotes) {
       if (all_of(VectorizableTree[Idx]->Scalars, [&](Value *V) {
-            uint32_t OrigBitWidth = DL->getTypeSizeInBits(V->getType());
+            uint32_t OrigBitWidth =
+                DL->getTypeSizeInBits(V->getType()->getScalarType());
             if (OrigBitWidth > MaxBitWidth) {
               APInt Mask = APInt::getBitsSetFrom(OrigBitWidth, MaxBitWidth);
               return MaskedValueIsZero(V, Mask, SimplifyQuery(*DL));

llvm/test/Transforms/SLPVectorizer/revec.ll

@@ -222,3 +222,19 @@ entry:
   %cmp = icmp ugt <4 x i32> %vbsl, <i32 2, i32 3, i32 4, i32 5>
   ret <4 x i1> %cmp
 }
+
+define void @test7() {
+; CHECK-LABEL: @test7(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i64> @llvm.vector.insert.v16i64.v8i64(<16 x i64> poison, <8 x i64> zeroinitializer, i64 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = call <16 x i64> @llvm.vector.insert.v16i64.v8i64(<16 x i64> [[TMP1]], <8 x i64> zeroinitializer, i64 8)
+; CHECK-NEXT:    [[TMP3:%.*]] = trunc <16 x i64> [[TMP2]] to <16 x i16>
+; CHECK-NEXT:    store <16 x i16> [[TMP3]], ptr null, align 2
+; CHECK-NEXT:    ret void
+;
+  %1 = getelementptr i8, ptr null, i64 16
+  %2 = trunc <8 x i64> zeroinitializer to <8 x i16>
+  store <8 x i16> %2, ptr %1, align 2
+  %3 = trunc <8 x i64> zeroinitializer to <8 x i16>
+  store <8 x i16> %3, ptr null, align 2
+  ret void
+}