[VP][RISCV] Add a vp.load.ff intrinsic for fault only first load.

llvm/docs/LangRef.rst

@@ -23943,6 +23943,63 @@ Examples:
      %also.r = call <8 x i8> @llvm.masked.load.v8i8.p0(ptr %ptr, i32 2, <8 x i1> %mask, <8 x i8> poison)
 
 
+.. _int_vp_ff_load:
+
+'``llvm.vp.ff.load``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Syntax:
+"""""""
+This is an overloaded intrinsic.
+
+::
+
+    declare {<4 x float>, i32} @llvm.vp.load.ff.v4f32.p0(ptr %ptr, <4 x i1> %mask, i32 %evl)
+    declare {<vscale x 2 x i16>, i32} @llvm.vp.load.ff.nxv2i16.p0(ptr %ptr, <vscale x 2 x i1> %mask, i32 %evl)
+    declare {<8 x float>, i32} @llvm.vp.load.ff.v8f32.p1(ptr addrspace(1) %ptr, <8 x i1> %mask, i32 %evl)
+    declare {<vscale x 1 x i64>, i32} @llvm.vp.load.ff.nxv1i64.p6(ptr addrspace(6) %ptr, <vscale x 1 x i1> %mask, i32 %evl)
+
+Overview:
+"""""""""
+
+The '``llvm.vp.load.ff.*``' intrinsic is similar to '``llvm.vp.load.*``', but
+will not trap if there are not ``evl`` readable elements at the pointer.
+
+Arguments:
+""""""""""
+
+The first argument is the base pointer for the load. The second argument is a
+vector of boolean values with the same number of elements as the first return
+type.  The third is the explicit vector length of the operation. The first
+return type and underlying type of the base pointer are the same vector types.
+
+The :ref:`align <attr_align>` parameter attribute can be provided for the first
+argument.
+
+Semantics:
+""""""""""
+
+The '``llvm.vp.load.ff``' intrinsic reads a vector from memory similar to
+'``llvm.vp.load``, but will only trap if the first lane is unreadable. If
+any other lane is unreadable, the number of successfully read lanes will
+be returned in the second return value. The result in the first return value
+for the lanes that were not successfully read is
+:ref:`poison value <poisonvalues>`. If ``evl`` is 0, no read occurs and thus no
+trap can occur for the first lane. If ``mask`` is 0 for the first lane, no
+trap occurs. This intrinsic is allowed to read fewer than ``evl`` lanes even
+if no trap would occur. If ``evl`` is non-zero, the result in the second result
+must be at least 1 even if the first lane is disabled by ``mask``.
+
+The default alignment is taken as the ABI alignment of the first return
+type as specified by the :ref:`datalayout string<langref_datalayout>`.
+
+Examples:
+"""""""""
+
+.. code-block:: text
+
+     %r = call {<8 x i8>, i32} @llvm.vp.load.ff.v8i8.p0(ptr align 2 %ptr, <8 x i1> %mask, i32 %evl)
+
 .. _int_vp_store:
 
 '``llvm.vp.store``' Intrinsic

llvm/include/llvm/CodeGen/SelectionDAG.h

@@ -1572,6 +1572,8 @@ class SelectionDAG {
   SDValue getMaskedHistogram(SDVTList VTs, EVT MemVT, const SDLoc &dl,
                              ArrayRef<SDValue> Ops, MachineMemOperand *MMO,
                              ISD::MemIndexType IndexType);
+  SDValue getLoadFFVP(EVT VT, const SDLoc &dl, SDValue Chain, SDValue Ptr,
+                      SDValue Mask, SDValue EVL, MachineMemOperand *MMO);
 
   SDValue getGetFPEnv(SDValue Chain, const SDLoc &dl, SDValue Ptr, EVT MemVT,
                       MachineMemOperand *MMO);

llvm/include/llvm/CodeGen/SelectionDAGNodes.h

@@ -3057,6 +3057,23 @@ class MaskedHistogramSDNode : public MaskedGatherScatterSDNode {
   }
 };
 
+class VPLoadFFSDNode : public MemSDNode {
+public:
+  friend class SelectionDAG;
+
+  VPLoadFFSDNode(unsigned Order, const DebugLoc &dl, SDVTList VTs, EVT MemVT,
+                 MachineMemOperand *MMO)
+      : MemSDNode(ISD::VP_LOAD_FF, Order, dl, VTs, MemVT, MMO) {}
+
+  const SDValue &getBasePtr() const { return getOperand(1); }
+  const SDValue &getMask() const { return getOperand(2); }
+  const SDValue &getVectorLength() const { return getOperand(3); }
+
+  static bool classof(const SDNode *N) {
+    return N->getOpcode() == ISD::VP_LOAD_FF;
+  }
+};
+
 class FPStateAccessSDNode : public MemSDNode {
 public:
   friend class SelectionDAG;

llvm/include/llvm/IR/Intrinsics.td

@@ -1912,6 +1912,12 @@ def int_vp_load  : DefaultAttrsIntrinsic<[ llvm_anyvector_ty],
                                llvm_i32_ty],
                              [ NoCapture<ArgIndex<0>>, IntrNoSync, IntrReadMem, IntrWillReturn, IntrArgMemOnly ]>;
 
+def int_vp_load_ff : DefaultAttrsIntrinsic<[ llvm_anyvector_ty, llvm_i32_ty ],
+                             [ llvm_anyptr_ty,
+                               LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                               llvm_i32_ty],
+                             [ NoCapture<ArgIndex<0>>, IntrNoSync, IntrReadMem, IntrWillReturn, IntrArgMemOnly ]>;
+
 def int_vp_gather: DefaultAttrsIntrinsic<[ llvm_anyvector_ty],
                              [ LLVMVectorOfAnyPointersToElt<0>,
                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,

llvm/include/llvm/IR/VPIntrinsics.def

@@ -587,6 +587,12 @@ VP_PROPERTY_FUNCTIONAL_OPC(Load)
 VP_PROPERTY_FUNCTIONAL_INTRINSIC(masked_load)
 END_REGISTER_VP(vp_load, VP_LOAD)
 
+BEGIN_REGISTER_VP_INTRINSIC(vp_load_ff, 1, 2)
+// val,chain = VP_LOAD_FF chain,base,mask,evl
+BEGIN_REGISTER_VP_SDNODE(VP_LOAD_FF, -1, vp_load_ff, 2, 3)
+HELPER_MAP_VPID_TO_VPSD(vp_load_ff, VP_LOAD_FF)
+VP_PROPERTY_NO_FUNCTIONAL
+END_REGISTER_VP(vp_load_ff, VP_LOAD_FF)
 // llvm.experimental.vp.strided.load(ptr,stride,mask,vlen)
 BEGIN_REGISTER_VP_INTRINSIC(experimental_vp_strided_load, 2, 3)
 // chain = EXPERIMENTAL_VP_STRIDED_LOAD chain,base,offset,stride,mask,evl

llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h

@@ -958,6 +958,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   void SplitVecRes_INSERT_VECTOR_ELT(SDNode *N, SDValue &Lo, SDValue &Hi);
   void SplitVecRes_LOAD(LoadSDNode *LD, SDValue &Lo, SDValue &Hi);
   void SplitVecRes_VP_LOAD(VPLoadSDNode *LD, SDValue &Lo, SDValue &Hi);
+  void SplitVecRes_VP_LOAD_FF(VPLoadFFSDNode *LD, SDValue &Lo, SDValue &Hi);
   void SplitVecRes_VP_STRIDED_LOAD(VPStridedLoadSDNode *SLD, SDValue &Lo,
                                    SDValue &Hi);
   void SplitVecRes_MLOAD(MaskedLoadSDNode *MLD, SDValue &Lo, SDValue &Hi);
@@ -1060,6 +1061,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue WidenVecRes_INSERT_VECTOR_ELT(SDNode* N);
   SDValue WidenVecRes_LOAD(SDNode* N);
   SDValue WidenVecRes_VP_LOAD(VPLoadSDNode *N);
+  SDValue WidenVecRes_VP_LOAD_FF(VPLoadFFSDNode *N);
   SDValue WidenVecRes_VP_STRIDED_LOAD(VPStridedLoadSDNode *N);
   SDValue WidenVecRes_VECTOR_COMPRESS(SDNode *N);
   SDValue WidenVecRes_MLOAD(MaskedLoadSDNode* N);

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

@@ -1163,6 +1163,9 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::VP_LOAD:
     SplitVecRes_VP_LOAD(cast<VPLoadSDNode>(N), Lo, Hi);
     break;
+  case ISD::VP_LOAD_FF:
+    SplitVecRes_VP_LOAD_FF(cast<VPLoadFFSDNode>(N), Lo, Hi);
+    break;
   case ISD::EXPERIMENTAL_VP_STRIDED_LOAD:
     SplitVecRes_VP_STRIDED_LOAD(cast<VPStridedLoadSDNode>(N), Lo, Hi);
     break;
@@ -2232,6 +2235,53 @@ void DAGTypeLegalizer::SplitVecRes_VP_LOAD(VPLoadSDNode *LD, SDValue &Lo,
   ReplaceValueWith(SDValue(LD, 1), Ch);
 }
 
+void DAGTypeLegalizer::SplitVecRes_VP_LOAD_FF(VPLoadFFSDNode *LD, SDValue &Lo,
+                                              SDValue &Hi) {
+  EVT LoVT, HiVT;
+  SDLoc dl(LD);
+  std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(LD->getValueType(0));
+
+  SDValue Ch = LD->getChain();
+  SDValue Ptr = LD->getBasePtr();
+  Align Alignment = LD->getOriginalAlign();
+  SDValue Mask = LD->getMask();
+  SDValue EVL = LD->getVectorLength();
+  EVT MemoryVT = LD->getMemoryVT();
+
+  EVT LoMemVT, HiMemVT;
+  bool HiIsEmpty = false;
+  std::tie(LoMemVT, HiMemVT) =
+      DAG.GetDependentSplitDestVTs(MemoryVT, LoVT, &HiIsEmpty);
+
+  // Split Mask operand
+  SDValue MaskLo, MaskHi;
+  if (Mask.getOpcode() == ISD::SETCC) {
+    SplitVecRes_SETCC(Mask.getNode(), MaskLo, MaskHi);
+  } else {
+    if (getTypeAction(Mask.getValueType()) == TargetLowering::TypeSplitVector)
+      GetSplitVector(Mask, MaskLo, MaskHi);
+    else
+      std::tie(MaskLo, MaskHi) = DAG.SplitVector(Mask, dl);
+  }
+
+  // Split EVL operand
+  SDValue EVLLo, EVLHi;
+  std::tie(EVLLo, EVLHi) = DAG.SplitEVL(EVL, LD->getValueType(0), dl);
+
+  MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
+      LD->getPointerInfo(), MachineMemOperand::MOLoad,
+      LocationSize::beforeOrAfterPointer(), Alignment, LD->getAAInfo(),
+      LD->getRanges());
+
+  Lo = DAG.getLoadFFVP(LoVT, dl, Ch, Ptr, MaskLo, EVLLo, MMO);
+
+  // Fill the upper half with poison.
+  Hi = DAG.getUNDEF(HiVT);
+
+  ReplaceValueWith(SDValue(LD, 1), Lo.getValue(1));
+  ReplaceValueWith(SDValue(LD, 2), Lo.getValue(2));
+}
+
 void DAGTypeLegalizer::SplitVecRes_VP_STRIDED_LOAD(VPStridedLoadSDNode *SLD,
                                                    SDValue &Lo, SDValue &Hi) {
   assert(SLD->isUnindexed() &&
@@ -4599,6 +4649,9 @@ void DAGTypeLegalizer::WidenVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::VP_LOAD:
     Res = WidenVecRes_VP_LOAD(cast<VPLoadSDNode>(N));
     break;
+  case ISD::VP_LOAD_FF:
+    Res = WidenVecRes_VP_LOAD_FF(cast<VPLoadFFSDNode>(N));
+    break;
   case ISD::EXPERIMENTAL_VP_STRIDED_LOAD:
     Res = WidenVecRes_VP_STRIDED_LOAD(cast<VPStridedLoadSDNode>(N));
     break;
@@ -6063,6 +6116,29 @@ SDValue DAGTypeLegalizer::WidenVecRes_VP_LOAD(VPLoadSDNode *N) {
   return Res;
 }
 
+SDValue DAGTypeLegalizer::WidenVecRes_VP_LOAD_FF(VPLoadFFSDNode *N) {
+  EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+  SDValue Mask = N->getMask();
+  SDValue EVL = N->getVectorLength();
+  SDLoc dl(N);
+
+  // The mask should be widened as well
+  assert(getTypeAction(Mask.getValueType()) ==
+             TargetLowering::TypeWidenVector &&
+         "Unable to widen binary VP op");
+  Mask = GetWidenedVector(Mask);
+  assert(Mask.getValueType().getVectorElementCount() ==
+             TLI.getTypeToTransformTo(*DAG.getContext(), Mask.getValueType())
+                 .getVectorElementCount() &&
+         "Unable to widen vector load");
+
+  SDValue Res = DAG.getLoadFFVP(WidenVT, dl, N->getChain(), N->getBasePtr(),
+                                Mask, EVL, N->getMemOperand());
+  ReplaceValueWith(SDValue(N, 1), Res.getValue(1));
+  ReplaceValueWith(SDValue(N, 2), Res.getValue(2));
+  return Res;
+}
+
 SDValue DAGTypeLegalizer::WidenVecRes_VP_STRIDED_LOAD(VPStridedLoadSDNode *N) {
   SDLoc DL(N);
 

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

@@ -846,6 +846,14 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) {
     ID.AddInteger(ELD->getMemOperand()->getFlags());
     break;
   }
+  case ISD::VP_LOAD_FF: {
+    const VPLoadFFSDNode *LD = cast<VPLoadFFSDNode>(N);
+    ID.AddInteger(LD->getMemoryVT().getRawBits());
+    ID.AddInteger(LD->getRawSubclassData());
+    ID.AddInteger(LD->getPointerInfo().getAddrSpace());
+    ID.AddInteger(LD->getMemOperand()->getFlags());
+    break;
+  }
   case ISD::VP_STORE: {
     const VPStoreSDNode *EST = cast<VPStoreSDNode>(N);
     ID.AddInteger(EST->getMemoryVT().getRawBits());
@@ -10123,6 +10131,34 @@ SDValue SelectionDAG::getMaskedHistogram(SDVTList VTs, EVT MemVT,
   return V;
 }
 
+SDValue SelectionDAG::getLoadFFVP(EVT VT, const SDLoc &dl, SDValue Chain,
+                                  SDValue Ptr, SDValue Mask, SDValue EVL,
+                                  MachineMemOperand *MMO) {
+  SDVTList VTs = getVTList(VT, EVL.getValueType(), MVT::Other);
+  SDValue Ops[] = {Chain, Ptr, Mask, EVL};
+  FoldingSetNodeID ID;
+  AddNodeIDNode(ID, ISD::VP_LOAD_FF, VTs, Ops);
+  ID.AddInteger(VT.getRawBits());
+  ID.AddInteger(getSyntheticNodeSubclassData<VPLoadFFSDNode>(dl.getIROrder(),
+                                                             VTs, VT, MMO));
+  ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
+  ID.AddInteger(MMO->getFlags());
+  void *IP = nullptr;
+  if (SDNode *E = FindNodeOrInsertPos(ID, dl, IP)) {
+    cast<VPLoadFFSDNode>(E)->refineAlignment(MMO);
+    return SDValue(E, 0);
+  }
+  auto *N = newSDNode<VPLoadFFSDNode>(dl.getIROrder(), dl.getDebugLoc(), VTs,
+                                      VT, MMO);
+  createOperands(N, Ops);
+
+  CSEMap.InsertNode(N, IP);
+  InsertNode(N);
+  SDValue V(N, 0);
+  NewSDValueDbgMsg(V, "Creating new node: ", this);
+  return V;
+}
+
 SDValue SelectionDAG::getGetFPEnv(SDValue Chain, const SDLoc &dl, SDValue Ptr,
                                   EVT MemVT, MachineMemOperand *MMO) {
   assert(Chain.getValueType() == MVT::Other && "Invalid chain type");

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -8473,6 +8473,35 @@ void SelectionDAGBuilder::visitVPLoad(
   setValue(&VPIntrin, LD);
 }
 
+void SelectionDAGBuilder::visitVPLoadFF(
+    const VPIntrinsic &VPIntrin, EVT VT, EVT EVLVT,
+    const SmallVectorImpl<SDValue> &OpValues) {
+  assert(OpValues.size() == 3);
+  SDLoc DL = getCurSDLoc();
+  Value *PtrOperand = VPIntrin.getArgOperand(0);
+  MaybeAlign Alignment = VPIntrin.getPointerAlignment();
+  AAMDNodes AAInfo = VPIntrin.getAAMetadata();
+  const MDNode *Ranges = VPIntrin.getMetadata(LLVMContext::MD_range);
+  SDValue LD;
+  bool AddToChain = true;
+  // Do not serialize variable-length loads of constant memory with
+  // anything.
+  if (!Alignment)
+    Alignment = DAG.getEVTAlign(VT);
+  MemoryLocation ML = MemoryLocation::getAfter(PtrOperand, AAInfo);
+  AddToChain = !BatchAA || !BatchAA->pointsToConstantMemory(ML);
+  SDValue InChain = AddToChain ? DAG.getRoot() : DAG.getEntryNode();
+  MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
+      MachinePointerInfo(PtrOperand), MachineMemOperand::MOLoad,
+      LocationSize::beforeOrAfterPointer(), *Alignment, AAInfo, Ranges);
+  LD = DAG.getLoadFFVP(VT, DL, InChain, OpValues[0], OpValues[1], OpValues[2],
+                       MMO);
+  SDValue Trunc = DAG.getNode(ISD::TRUNCATE, DL, EVLVT, LD.getValue(1));
+  if (AddToChain)
+    PendingLoads.push_back(LD.getValue(2));
+  setValue(&VPIntrin, DAG.getMergeValues({LD.getValue(0), Trunc}, DL));
+}
+
 void SelectionDAGBuilder::visitVPGather(
     const VPIntrinsic &VPIntrin, EVT VT,
     const SmallVectorImpl<SDValue> &OpValues) {
@@ -8706,6 +8735,9 @@ void SelectionDAGBuilder::visitVectorPredicationIntrinsic(
   case ISD::VP_LOAD:
     visitVPLoad(VPIntrin, ValueVTs[0], OpValues);
     break;
+  case ISD::VP_LOAD_FF:
+    visitVPLoadFF(VPIntrin, ValueVTs[0], ValueVTs[1], OpValues);
+    break;
   case ISD::VP_GATHER:
     visitVPGather(VPIntrin, ValueVTs[0], OpValues);
     break;

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

@@ -632,6 +632,8 @@ class SelectionDAGBuilder {
   void visitVectorExtractLastActive(const CallInst &I, unsigned Intrinsic);
   void visitVPLoad(const VPIntrinsic &VPIntrin, EVT VT,
                    const SmallVectorImpl<SDValue> &OpValues);
+  void visitVPLoadFF(const VPIntrinsic &VPIntrin, EVT VT, EVT EVLVT,
+                     const SmallVectorImpl<SDValue> &OpValues);
   void visitVPStore(const VPIntrinsic &VPIntrin,
                     const SmallVectorImpl<SDValue> &OpValues);
   void visitVPGather(const VPIntrinsic &VPIntrin, EVT VT,

llvm/lib/IR/IntrinsicInst.cpp

@@ -448,6 +448,7 @@ VPIntrinsic::getMemoryPointerParamPos(Intrinsic::ID VPID) {
   case Intrinsic::experimental_vp_strided_store:
     return 1;
   case Intrinsic::vp_load:
+  case Intrinsic::vp_load_ff:
   case Intrinsic::vp_gather:
   case Intrinsic::experimental_vp_strided_load:
     return 0;
@@ -671,6 +672,10 @@ Function *VPIntrinsic::getOrInsertDeclarationForParams(
     VPFunc = Intrinsic::getOrInsertDeclaration(
         M, VPID, {ReturnType, Params[0]->getType()});
     break;
+  case Intrinsic::vp_load_ff:
+    VPFunc = Intrinsic::getOrInsertDeclaration(
+        M, VPID, {ReturnType->getStructElementType(0), Params[0]->getType()});
+    break;
   case Intrinsic::experimental_vp_strided_load:
     VPFunc = Intrinsic::getOrInsertDeclaration(
         M, VPID, {ReturnType, Params[0]->getType(), Params[1]->getType()});