[SLP][REVEC] Make reorderTopToBottom support ShuffleVectorInst. (#117310)

We don't want reorderTopToBottom to reorder ShuffleVectorInst (because
ShuffleVectorInst currently supports only a limited set of patterns).
Either we make ShuffleVectorInst support more patterns, or we let
ReorderIndices reorder the result of the vectorization of
ShuffleVectorInst. We choose the latter solution.

Author: HanKuanChen

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -6017,8 +6017,9 @@ void BoUpSLP::reorderTopToBottom() {
       }
       if ((TE->State == TreeEntry::Vectorize ||
            TE->State == TreeEntry::StridedVectorize) &&
-          isa<ExtractElementInst, ExtractValueInst, LoadInst, StoreInst,
-              InsertElementInst>(TE->getMainOp())) {
+          (isa<ExtractElementInst, ExtractValueInst, LoadInst, StoreInst,
+               InsertElementInst>(TE->getMainOp()) ||
+           (SLPReVec && isa<ShuffleVectorInst>(TE->getMainOp())))) {
         assert(!TE->isAltShuffle() &&
                "Alternate instructions are only supported by BinaryOperator "
                "and CastInst.");

llvm/test/Transforms/SLPVectorizer/revec.ll

@@ -409,3 +409,41 @@ entry:
   %23 = fcmp ogt <8 x float> zeroinitializer, %19
   ret void
 }
+
+define void @test13(<8 x i32> %0, ptr %out0, ptr %out1, ptr %out2) {
+; CHECK-LABEL: @test13(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i32> @llvm.vector.insert.v32i32.v8i32(<32 x i32> poison, <8 x i32> [[TMP0:%.*]], i64 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <32 x i32> [[TMP1]], <32 x i32> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <32 x i32> [[TMP1]], <32 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <32 x i32> [[TMP1]], <32 x i32> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    br label [[FOR_END_LOOPEXIT:%.*]]
+; CHECK:       for.end.loopexit:
+; CHECK-NEXT:    [[TMP5:%.*]] = phi <16 x i32> [ [[TMP4]], [[ENTRY:%.*]] ]
+; CHECK-NEXT:    [[TMP6:%.*]] = call <4 x i32> @llvm.vector.extract.v4i32.v16i32(<16 x i32> [[TMP5]], i64 12)
+; CHECK-NEXT:    [[OR0:%.*]] = or <4 x i32> [[TMP6]], zeroinitializer
+; CHECK-NEXT:    store <4 x i32> [[OR0]], ptr [[OUT0:%.*]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = call <4 x i32> @llvm.vector.extract.v4i32.v16i32(<16 x i32> [[TMP4]], i64 0)
+; CHECK-NEXT:    store <4 x i32> [[TMP7]], ptr [[OUT1:%.*]], align 4
+; CHECK-NEXT:    [[TMP8:%.*]] = call <4 x i32> @llvm.vector.extract.v4i32.v16i32(<16 x i32> [[TMP4]], i64 8)
+; CHECK-NEXT:    store <4 x i32> [[TMP8]], ptr [[OUT2:%.*]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %1 = shufflevector <8 x i32> %0, <8 x i32> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %2 = shufflevector <8 x i32> %0, <8 x i32> zeroinitializer, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  %3 = shufflevector <8 x i32> %0, <8 x i32> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = shufflevector <8 x i32> %0, <8 x i32> zeroinitializer, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  br label %for.end.loopexit
+
+for.end.loopexit:
+  %phi0 = phi <4 x i32> [ %1, %entry ]
+  %phi1 = phi <4 x i32> [ %2, %entry ]
+  %phi2 = phi <4 x i32> [ %3, %entry ]
+  %phi3 = phi <4 x i32> [ %4, %entry ]
+  %or0 = or <4 x i32> %phi1, zeroinitializer
+  store <4 x i32> %or0, ptr %out0, align 4
+  store <4 x i32> %1, ptr %out1, align 4
+  store <4 x i32> %4, ptr %out2, align 4
+  ret void
+}