update uses in SLPVectorizer

- pass down the previously defined TTI

Author: inbelic

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -1091,7 +1091,8 @@ static bool allSameType(ArrayRef<Value *> VL) {
 /// \returns True if in-tree use also needs extract. This refers to
 /// possible scalar operand in vectorized instruction.
 static bool doesInTreeUserNeedToExtract(Value *Scalar, Instruction *UserInst,
-                                        TargetLibraryInfo *TLI) {
+                                        TargetLibraryInfo *TLI,
+                                        const TargetTransformInfo *TTI) {
   if (!UserInst)
     return false;
   unsigned Opcode = UserInst->getOpcode();
@@ -1108,7 +1109,7 @@ static bool doesInTreeUserNeedToExtract(Value *Scalar, Instruction *UserInst,
     CallInst *CI = cast<CallInst>(UserInst);
     Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
     return any_of(enumerate(CI->args()), [&](auto &&Arg) {
-      return isVectorIntrinsicWithScalarOpAtArg(ID, Arg.index()) &&
+      return isVectorIntrinsicWithScalarOpAtArg(ID, Arg.index(), TTI) &&
              Arg.value().get() == Scalar;
     });
   }
@@ -6505,7 +6506,7 @@ void BoUpSLP::buildExternalUses(
           // be used.
           if (UseEntry->State == TreeEntry::ScatterVectorize ||
               !doesInTreeUserNeedToExtract(
-                  Scalar, getRootEntryInstruction(*UseEntry), TLI)) {
+                  Scalar, getRootEntryInstruction(*UseEntry), TLI, TTI)) {
             LLVM_DEBUG(dbgs() << "SLP: \tInternal user will be removed:" << *U
                               << ".\n");
             assert(!UseEntry->isGather() && "Bad state");
@@ -7830,7 +7831,7 @@ BoUpSLP::TreeEntry::EntryState BoUpSLP::getScalarsVectorizationState(
     unsigned NumArgs = CI->arg_size();
     SmallVector<Value *, 4> ScalarArgs(NumArgs, nullptr);
     for (unsigned J = 0; J != NumArgs; ++J)
-      if (isVectorIntrinsicWithScalarOpAtArg(ID, J))
+      if (isVectorIntrinsicWithScalarOpAtArg(ID, J, TTI))
         ScalarArgs[J] = CI->getArgOperand(J);
     for (Value *V : VL) {
       CallInst *CI2 = dyn_cast<CallInst>(V);
@@ -7846,7 +7847,7 @@ BoUpSLP::TreeEntry::EntryState BoUpSLP::getScalarsVectorizationState(
       // Some intrinsics have scalar arguments and should be same in order for
       // them to be vectorized.
       for (unsigned J = 0; J != NumArgs; ++J) {
-        if (isVectorIntrinsicWithScalarOpAtArg(ID, J)) {
+        if (isVectorIntrinsicWithScalarOpAtArg(ID, J, TTI)) {
           Value *A1J = CI2->getArgOperand(J);
           if (ScalarArgs[J] != A1J) {
             LLVM_DEBUG(dbgs()
@@ -8718,7 +8719,7 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
       for (unsigned I : seq<unsigned>(CI->arg_size())) {
         // For scalar operands no need to create an entry since no need to
         // vectorize it.
-        if (isVectorIntrinsicWithScalarOpAtArg(ID, I))
+        if (isVectorIntrinsicWithScalarOpAtArg(ID, I, TTI))
           continue;
         buildTree_rec(TE->getOperand(I), Depth + 1, {TE, I});
       }
@@ -10948,14 +10949,14 @@ TTI::CastContextHint BoUpSLP::getCastContextHint(const TreeEntry &TE) const {
 
 /// Builds the arguments types vector for the given call instruction with the
 /// given \p ID for the specified vector factor.
-static SmallVector<Type *> buildIntrinsicArgTypes(const CallInst *CI,
-                                                  const Intrinsic::ID ID,
-                                                  const unsigned VF,
-                                                  unsigned MinBW) {
+static SmallVector<Type *>
+buildIntrinsicArgTypes(const CallInst *CI, const Intrinsic::ID ID,
+                       const unsigned VF, unsigned MinBW,
+                       const TargetTransformInfo *TTI) {
   SmallVector<Type *> ArgTys;
   for (auto [Idx, Arg] : enumerate(CI->args())) {
     if (ID != Intrinsic::not_intrinsic) {
-      if (isVectorIntrinsicWithScalarOpAtArg(ID, Idx)) {
+      if (isVectorIntrinsicWithScalarOpAtArg(ID, Idx, TTI)) {
         ArgTys.push_back(Arg->getType());
         continue;
       }
@@ -11657,9 +11658,9 @@ BoUpSLP::getEntryCost(const TreeEntry *E, ArrayRef<Value *> VectorizedVals,
     auto GetVectorCost = [=](InstructionCost CommonCost) {
       auto *CI = cast<CallInst>(VL0);
       Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
-      SmallVector<Type *> ArgTys =
-          buildIntrinsicArgTypes(CI, ID, VecTy->getNumElements(),
-                                 It != MinBWs.end() ? It->second.first : 0);
+      SmallVector<Type *> ArgTys = buildIntrinsicArgTypes(
+          CI, ID, VecTy->getNumElements(),
+          It != MinBWs.end() ? It->second.first : 0, TTI);
       auto VecCallCosts = getVectorCallCosts(CI, VecTy, TTI, TLI, ArgTys);
       return std::min(VecCallCosts.first, VecCallCosts.second) + CommonCost;
     };
@@ -15817,9 +15818,9 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
 
       Intrinsic::ID ID = getVectorIntrinsicIDForCall(CI, TLI);
 
-      SmallVector<Type *> ArgTys =
-          buildIntrinsicArgTypes(CI, ID, VecTy->getNumElements(),
-                                 It != MinBWs.end() ? It->second.first : 0);
+      SmallVector<Type *> ArgTys = buildIntrinsicArgTypes(
+          CI, ID, VecTy->getNumElements(),
+          It != MinBWs.end() ? It->second.first : 0, TTI);
       auto VecCallCosts = getVectorCallCosts(CI, VecTy, TTI, TLI, ArgTys);
       bool UseIntrinsic = ID != Intrinsic::not_intrinsic &&
                           VecCallCosts.first <= VecCallCosts.second;
@@ -15835,7 +15836,7 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
         ValueList OpVL;
         // Some intrinsics have scalar arguments. This argument should not be
         // vectorized.
-        if (UseIntrinsic && isVectorIntrinsicWithScalarOpAtArg(ID, I)) {
+        if (UseIntrinsic && isVectorIntrinsicWithScalarOpAtArg(ID, I, TTI)) {
           ScalarArg = CEI->getArgOperand(I);
           // if decided to reduce bitwidth of abs intrinsic, it second argument
           // must be set false (do not return poison, if value issigned min).
@@ -16374,7 +16375,7 @@ BoUpSLP::vectorizeTree(const ExtraValueToDebugLocsMap &ExternallyUsedValues,
                                E->State == TreeEntry::StridedVectorize) &&
                               doesInTreeUserNeedToExtract(
                                   Scalar, getRootEntryInstruction(*UseEntry),
-                                  TLI);
+                                  TLI, TTI);
                      })) &&
              "Scalar with nullptr User must be registered in "
              "ExternallyUsedValues map or remain as scalar in vectorized "
@@ -17870,7 +17871,8 @@ bool BoUpSLP::collectValuesToDemote(
     // Choose the best bitwidth based on cost estimations.
     auto Checker = [&](unsigned BitWidth, unsigned) {
       unsigned MinBW = PowerOf2Ceil(BitWidth);
-      SmallVector<Type *> ArgTys = buildIntrinsicArgTypes(IC, ID, VF, MinBW);
+      SmallVector<Type *> ArgTys =
+          buildIntrinsicArgTypes(IC, ID, VF, MinBW, TTI);
       auto VecCallCosts = getVectorCallCosts(
           IC, getWidenedType(IntegerType::get(IC->getContext(), MinBW), VF),
           TTI, TLI, ArgTys);