[llvm] [LV] Teach the vectorizer to cost and vectorize llvm.sincos intrinsics (PR #123210)

[Benjamin Maxwell via llvm-commits]

================
@@ -285,6 +286,64 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
     return false;
   }
 
+  /// Several intrinsics struct-ret (including llvm.sincos[pi] and llvm.modf)
+  /// can be lowered to a vector library call (for certain VFs). The vector
+  /// library functions correspond to the scalar calls (e.g. sincos or modf),
+  /// which unlike the intrinsic return values via output pointers. This helper
+  /// checks if a vector call exists for the given intrinsic, and returns the
+  /// cost, which includes the cost of the mask (if required), and the loads for
+  /// values returned via output pointers. \p LC is the scalar libcall and
+  /// \p CallRetElementIndex (optional) is the struct element which is mapped to
+  /// the call return value. If std::nullopt is returned, the no vector library
+  /// call is available, so the intrinsic should be assigned the default cost
+  /// (e.g. scalarization).
+  std::optional<InstructionCost> getMultipleResultIntrinsicVectorLibCallCost(
+      const IntrinsicCostAttributes &ICA, TTI::TargetCostKind CostKind,
+      RTLIB::Libcall LC, std::optional<unsigned> CallRetElementIndex = {}) {
+    Type *RetTy = ICA.getReturnType();
+    // Vector variants of the intrinsic can be mapped to a vector library call.
+    auto const *LibInfo = ICA.getLibInfo();
+    if (!LibInfo || !isa<StructType>(RetTy) ||
+        !isVectorizedStructTy(cast<StructType>(RetTy)))
+      return std::nullopt;
+
+    // Find associated libcall.
+    const char *LCName = getTLI()->getLibcallName(LC);
+    if (!LC || !LCName)
+      return std::nullopt;
+
+    // Search for a corresponding vector variant.
+    LLVMContext &Ctx = RetTy->getContext();
+    ElementCount VF = getVectorizedTypeVF(RetTy);
+    VecDesc const *VD = nullptr;
+    for (bool Masked : {false, true}) {
+      if ((VD = LibInfo->getVectorMappingInfo(LCName, VF, Masked)))
+        break;
+    }
+    if (!VD)
+      return std::nullopt;
+
+    // Cost the call + mask.
+    auto Cost =
+        thisT()->getCallInstrCost(nullptr, RetTy, ICA.getArgTypes(), CostKind);
+    if (VD->isMasked())
+      Cost += thisT()->getShuffleCost(
+          TargetTransformInfo::SK_Broadcast,
+          VectorType::get(IntegerType::getInt1Ty(Ctx), VF), {}, CostKind, 0,
+          nullptr, {});
+
+    // Lowering to a library call (with output pointers) may require us to emit
----------------
MacDue wrote:

It is slightly pessimistic, but the cost is still reasonably low 13 (rather than 10 for a plain libcall), so the vectorizier still chooses to widen the intrinsic. It also means if libraries add structure-returning variants they'll preferred as they'll have a slightly lower cost.  

https://github.com/llvm/llvm-project/pull/123210