AMDGPU/R600: Remove MergeVectorStore

[jvesely]

Summary

This is handled by DAGCombiner in more generic way

[jvesely]

The downside is that char vectors are now broken in the same way as short vectors.

[arsenm]

LGTM, performStoreCombine is supposed to deal with this

llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h

 
 class AMDGPUTargetLowering : public TargetLowering {
 protected:
   const AMDGPUSubtarget *Subtarget;
 
   SDValue LowerEXTRACT_SUBVECTOR(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
-  /// \brief Lower vector stores by merging the vector elements into an integer
-  /// of the same bitwidth.
-  SDValue MergeVectorStore(const SDValue &Op, SelectionDAG &DAG) const;
   /// \brief Split a vector store into multiple scalar stores.
   /// \returns The resulting chain.
 
   SDValue LowerFREM(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFCEIL(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFTRUNC(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFRINT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFNEARBYINT(SDValue Op, SelectionDAG &DAG) const;

llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp

SDValue Ops[] = {
     DAG.getNode(ISD::CONCAT_VECTORS, SL, VT, LoLoad, HiLoad),
     DAG.getNode(ISD::TokenFactor, SL, MVT::Other,
                 LoLoad.getValue(1), HiLoad.getValue(1))
   };
 
   return DAG.getMergeValues(Ops, SL);
 }
 
-// FIXME: This isn't doing anything for SI. This should be used in a target
-// combine during type legalization.
-SDValue AMDGPUTargetLowering::MergeVectorStore(const SDValue &Op,
-                                               SelectionDAG &DAG) const {
-  StoreSDNode *Store = cast<StoreSDNode>(Op);
-  EVT MemVT = Store->getMemoryVT();
-  unsigned MemBits = MemVT.getSizeInBits();
-
-  // Byte stores are really expensive, so if possible, try to pack 32-bit vector
-  // truncating store into an i32 store.
-  // XXX: We could also handle optimize other vector bitwidths.
-  if (!MemVT.isVector() || MemBits > 32) {
-    return SDValue();
-  }
-
-  SDLoc DL(Op);
-  SDValue Value = Store->getValue();
-  EVT VT = Value.getValueType();
-  EVT ElemVT = VT.getVectorElementType();
-  SDValue Ptr = Store->getBasePtr();
-  EVT MemEltVT = MemVT.getVectorElementType();
-  unsigned MemEltBits = MemEltVT.getSizeInBits();
-  unsigned MemNumElements = MemVT.getVectorNumElements();
-  unsigned PackedSize = MemVT.getStoreSizeInBits();
-  SDValue Mask = DAG.getConstant((1 << MemEltBits) - 1, DL, MVT::i32);
-
-  assert(Value.getValueType().getScalarSizeInBits() >= 32);
-
-  SDValue PackedValue;
-  for (unsigned i = 0; i < MemNumElements; ++i) {
-    SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ElemVT, Value,
-                              DAG.getConstant(i, DL, MVT::i32));
-    Elt = DAG.getZExtOrTrunc(Elt, DL, MVT::i32);
-    Elt = DAG.getNode(ISD::AND, DL, MVT::i32, Elt, Mask); // getZeroExtendInReg
-
-    SDValue Shift = DAG.getConstant(MemEltBits * i, DL, MVT::i32);
-    Elt = DAG.getNode(ISD::SHL, DL, MVT::i32, Elt, Shift);
-
-    if (i == 0) {
-      PackedValue = Elt;
-    } else {
-      PackedValue = DAG.getNode(ISD::OR, DL, MVT::i32, PackedValue, Elt);
-    }
-  }
-
-  if (PackedSize < 32) {
-    EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), PackedSize);
-    return DAG.getTruncStore(Store->getChain(), DL, PackedValue, Ptr,
-                             Store->getMemOperand()->getPointerInfo(), PackedVT,
-                             Store->getAlignment(),
-                             Store->getMemOperand()->getFlags());
-  }
-
-  return DAG.getStore(Store->getChain(), DL, PackedValue, Ptr,
-                      Store->getMemOperand()->getPointerInfo(),
-                      Store->getAlignment(),
-                      Store->getMemOperand()->getFlags());
-}
-
 SDValue AMDGPUTargetLowering::SplitVectorStore(SDValue Op,
                                                SelectionDAG &DAG) const {
   StoreSDNode *Store = cast<StoreSDNode>(Op);
   SDValue Val = Store->getValue();
   EVT VT = Val.getValueType();
 
   // If this is a 2 element vector, we really want to scalarize and not create
   // weird 1 element vectors.

llvm/trunk/lib/Target/AMDGPU/R600ISelLowering.cpp

SDValue R600TargetLowering::lowerPrivateTruncStore(StoreSDNode *Store,
 
   SDValue Value = DAG.getNode(ISD::OR, DL, MVT::i32, Dst, ShiftedValue);
   return DAG.getNode(AMDGPUISD::REGISTER_STORE, DL, MVT::Other,
                      Chain, Value, Ptr,
                      DAG.getTargetConstant(0, DL, MVT::i32));
 }
 
 SDValue R600TargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
-  if (SDValue Result = AMDGPUTargetLowering::MergeVectorStore(Op, DAG))
-    return Result;
-
   StoreSDNode *StoreNode = cast<StoreSDNode>(Op);
   unsigned AS = StoreNode->getAddressSpace();
   SDValue Value = StoreNode->getValue();
   EVT ValueVT = Value.getValueType();
 
   if ((AS == AMDGPUAS::LOCAL_ADDRESS || AS == AMDGPUAS::PRIVATE_ADDRESS) &&
       ValueVT.isVector()) {
     return SplitVectorStore(Op, DAG);