[SLP]Fix PR59230: Use actual vector factor when sorting entries.

When we sort entries for attempting to reorder scalars, need to use
actual vectorization factor, not the number of scalars. Otherwise the
compiler crashes, if the scalars has to be reordered.

Differential Revision: https://reviews.llvm.org/D138819

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -4083,7 +4083,7 @@ void BoUpSLP::reorderTopToBottom() {
     SmallVector<OrdersType, 1> ExternalUserReorderIndices =
         findExternalStoreUsersReorderIndices(TE.get());
     if (!ExternalUserReorderIndices.empty()) {
-      VFToOrderedEntries[TE->Scalars.size()].insert(TE.get());
+      VFToOrderedEntries[TE->getVectorFactor()].insert(TE.get());
       ExternalUserReorderMap.try_emplace(TE.get(),
                                          std::move(ExternalUserReorderIndices));
     }
@@ -4103,7 +4103,7 @@ void BoUpSLP::reorderTopToBottom() {
           OpcodeMask.set(Lane);
       // If this pattern is supported by the target then we consider the order.
       if (TTIRef.isLegalAltInstr(VecTy, Opcode0, Opcode1, OpcodeMask)) {
-        VFToOrderedEntries[TE->Scalars.size()].insert(TE.get());
+        VFToOrderedEntries[TE->getVectorFactor()].insert(TE.get());
         AltShufflesToOrders.try_emplace(TE.get(), OrdersType());
       }
       // TODO: Check the reverse order too.
@@ -4141,7 +4141,7 @@ void BoUpSLP::reorderTopToBottom() {
   });
 
   // Reorder the graph nodes according to their vectorization factor.
-  for (unsigned VF = VectorizableTree.front()->Scalars.size(); VF > 1;
+  for (unsigned VF = VectorizableTree.front()->getVectorFactor(); VF > 1;
        VF /= 2) {
     auto It = VFToOrderedEntries.find(VF);
     if (It == VFToOrderedEntries.end())

llvm/test/Transforms/SLPVectorizer/X86/reorder-vf-to-resize.ll

@@ -0,0 +1,43 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=slp-vectorizer -mcpu=skx -mtriple=x86_64-unknown-linux-gnu -S < %s | FileCheck %s
+
+define void @main(ptr %0) {
+; CHECK-LABEL: @main(
+; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr [[TMP0:%.*]], align 8
+; CHECK-NEXT:    [[TMP3:%.*]] = fsub <2 x double> zeroinitializer, [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = fadd <2 x double> zeroinitializer, [[TMP2]]
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x double> [[TMP3]], <2 x double> [[TMP4]], <4 x i32> <i32 1, i32 2, i32 1, i32 2>
+; CHECK-NEXT:    [[TMP6:%.*]] = fmul <4 x double> [[TMP5]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = call <4 x double> @llvm.fabs.v4f64(<4 x double> [[TMP6]])
+; CHECK-NEXT:    [[TMP8:%.*]] = fcmp oeq <4 x double> [[TMP7]], zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[TMP8]])
+; CHECK-NEXT:    [[TMP10:%.*]] = select i1 [[TMP9]], double 0.000000e+00, double 0.000000e+00
+; CHECK-NEXT:    store double [[TMP10]], ptr null, align 8
+; CHECK-NEXT:    ret void
+;
+  %.unpack = load double, ptr %0, align 8
+  %.elt1 = getelementptr { double, double }, ptr %0, i64 0, i32 1
+  %.unpack2 = load double, ptr %.elt1, align 8
+  %2 = fadd double %.unpack, 0.000000e+00
+  %3 = fsub double 0.000000e+00, %.unpack2
+  %4 = fmul double %2, 0.000000e+00
+  %5 = call double @llvm.fabs.f64(double %4)
+  %6 = fmul double %3, 0.000000e+00
+  %7 = call double @llvm.fabs.f64(double %6)
+  %8 = fmul double %3, 0.000000e+00
+  %9 = call double @llvm.fabs.f64(double %8)
+  %10 = fmul double %2, 0.000000e+00
+  %11 = call double @llvm.fabs.f64(double %10)
+  %12 = fcmp oeq double %5, 0.000000e+00
+  %13 = fcmp oeq double %7, 0.000000e+00
+  %14 = or i1 %12, %13
+  %15 = fcmp oeq double %11, 0.000000e+00
+  %16 = or i1 %14, %15
+  %17 = fcmp oeq double %9, 0.000000e+00
+  %18 = or i1 %16, %17
+  %19 = select i1 %18, double 0.000000e+00, double 0.000000e+00
+  store double %19, ptr null, align 8
+  ret void
+}
+
+declare double @llvm.fabs.f64(double)