ISel: introduce vector ISD::LRINT, ISD::LLRINT; custom RISCV lowering (#66924)

The issue #55208 noticed that std::rint is vectorized by the
SLPVectorizer, but a very similar function, std::lrint, is not.
std::lrint corresponds to ISD::LRINT in the SelectionDAG, and
std::llrint is a familiar cousin corresponding to ISD::LLRINT. Now,
neither ISD::LRINT nor ISD::LLRINT have a corresponding vector variant,
and the LangRef makes this clear in the documentation of llvm.lrint.*
and llvm.llrint.*.

This patch extends the LangRef to include vector variants of
llvm.lrint.* and llvm.llrint.*, and lays the necessary ground-work of
scalarizing it for all targets. However, this patch would be devoid of
motivation unless we show the utility of these new vector variants.
Hence, the RISCV target has been chosen to implement a custom lowering
to the vfcvt.x.f.v instruction. The patch also includes a CostModel for
RISCV, and a trivial follow-up can potentially enable the SLPVectorizer
to vectorize std::lrint and std::llrint, fixing #55208.

The patch includes tests, obviously for the RISCV target, but also for
the X86, AArch64, and PowerPC targets to justify the addition of the
vector variants to the LangRef.

Author: artagnon

llvm/docs/LangRef.rst

@@ -15760,7 +15760,8 @@ Syntax:
 """""""
 
 This is an overloaded intrinsic. You can use ``llvm.lrint`` on any
-floating-point type. Not all targets support all types however.
+floating-point type or vector of floating-point type. Not all targets
+support all types however.
 
 ::
 
@@ -15804,7 +15805,8 @@ Syntax:
 """""""
 
 This is an overloaded intrinsic. You can use ``llvm.llrint`` on any
-floating-point type. Not all targets support all types however.
+floating-point type or vector of floating-point type. Not all targets
+support all types however.
 
 ::
 

llvm/include/llvm/CodeGen/BasicTTIImpl.h

@@ -1847,6 +1847,12 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
     case Intrinsic::rint:
       ISD = ISD::FRINT;
       break;
+    case Intrinsic::lrint:
+      ISD = ISD::LRINT;
+      break;
+    case Intrinsic::llrint:
+      ISD = ISD::LLRINT;
+      break;
     case Intrinsic::round:
       ISD = ISD::FROUND;
       break;

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -505,6 +505,7 @@ namespace {
     SDValue visitUINT_TO_FP(SDNode *N);
     SDValue visitFP_TO_SINT(SDNode *N);
     SDValue visitFP_TO_UINT(SDNode *N);
+    SDValue visitXRINT(SDNode *N);
     SDValue visitFP_ROUND(SDNode *N);
     SDValue visitFP_EXTEND(SDNode *N);
     SDValue visitFNEG(SDNode *N);
@@ -1911,6 +1912,7 @@ void DAGCombiner::Run(CombineLevel AtLevel) {
 }
 
 SDValue DAGCombiner::visit(SDNode *N) {
+  // clang-format off
   switch (N->getOpcode()) {
   default: break;
   case ISD::TokenFactor:        return visitTokenFactor(N);
@@ -2011,6 +2013,8 @@ SDValue DAGCombiner::visit(SDNode *N) {
   case ISD::UINT_TO_FP:         return visitUINT_TO_FP(N);
   case ISD::FP_TO_SINT:         return visitFP_TO_SINT(N);
   case ISD::FP_TO_UINT:         return visitFP_TO_UINT(N);
+  case ISD::LRINT:
+  case ISD::LLRINT:             return visitXRINT(N);
   case ISD::FP_ROUND:           return visitFP_ROUND(N);
   case ISD::FP_EXTEND:          return visitFP_EXTEND(N);
   case ISD::FNEG:               return visitFNEG(N);
@@ -2065,6 +2069,7 @@ SDValue DAGCombiner::visit(SDNode *N) {
 #include "llvm/IR/VPIntrinsics.def"
     return visitVPOp(N);
   }
+  // clang-format on
   return SDValue();
 }
 
@@ -17480,6 +17485,21 @@ SDValue DAGCombiner::visitFP_TO_UINT(SDNode *N) {
   return FoldIntToFPToInt(N, DAG);
 }
 
+SDValue DAGCombiner::visitXRINT(SDNode *N) {
+  SDValue N0 = N->getOperand(0);
+  EVT VT = N->getValueType(0);
+
+  // fold (lrint|llrint undef) -> undef
+  if (N0.isUndef())
+    return DAG.getUNDEF(VT);
+
+  // fold (lrint|llrint c1fp) -> c1
+  if (DAG.isConstantFPBuildVectorOrConstantFP(N0))
+    return DAG.getNode(N->getOpcode(), SDLoc(N), VT, N0);
+
+  return SDValue();
+}
+
 SDValue DAGCombiner::visitFP_ROUND(SDNode *N) {
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);

llvm/lib/CodeGen/SelectionDAG/LegalizeFloatTypes.cpp

@@ -2198,6 +2198,7 @@ bool DAGTypeLegalizer::PromoteFloatOperand(SDNode *N, unsigned OpNo) {
   // to use the promoted float operand.  Nodes that produce at least one
   // promotion-requiring floating point result have their operands legalized as
   // a part of PromoteFloatResult.
+  // clang-format off
   switch (N->getOpcode()) {
     default:
   #ifndef NDEBUG
@@ -2209,7 +2210,9 @@ bool DAGTypeLegalizer::PromoteFloatOperand(SDNode *N, unsigned OpNo) {
     case ISD::BITCAST:    R = PromoteFloatOp_BITCAST(N, OpNo); break;
     case ISD::FCOPYSIGN:  R = PromoteFloatOp_FCOPYSIGN(N, OpNo); break;
     case ISD::FP_TO_SINT:
-    case ISD::FP_TO_UINT: R = PromoteFloatOp_FP_TO_XINT(N, OpNo); break;
+    case ISD::FP_TO_UINT:
+    case ISD::LRINT:
+    case ISD::LLRINT:     R = PromoteFloatOp_UnaryOp(N, OpNo); break;
     case ISD::FP_TO_SINT_SAT:
     case ISD::FP_TO_UINT_SAT:
                           R = PromoteFloatOp_FP_TO_XINT_SAT(N, OpNo); break;
@@ -2218,6 +2221,7 @@ bool DAGTypeLegalizer::PromoteFloatOperand(SDNode *N, unsigned OpNo) {
     case ISD::SETCC:      R = PromoteFloatOp_SETCC(N, OpNo); break;
     case ISD::STORE:      R = PromoteFloatOp_STORE(N, OpNo); break;
   }
+  // clang-format on
 
   if (R.getNode())
     ReplaceValueWith(SDValue(N, 0), R);
@@ -2251,7 +2255,7 @@ SDValue DAGTypeLegalizer::PromoteFloatOp_FCOPYSIGN(SDNode *N, unsigned OpNo) {
 }
 
 // Convert the promoted float value to the desired integer type
-SDValue DAGTypeLegalizer::PromoteFloatOp_FP_TO_XINT(SDNode *N, unsigned OpNo) {
+SDValue DAGTypeLegalizer::PromoteFloatOp_UnaryOp(SDNode *N, unsigned OpNo) {
   SDValue Op = GetPromotedFloat(N->getOperand(0));
   return DAG.getNode(N->getOpcode(), SDLoc(N), N->getValueType(0), Op);
 }

llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h

@@ -711,7 +711,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue PromoteFloatOp_BITCAST(SDNode *N, unsigned OpNo);
   SDValue PromoteFloatOp_FCOPYSIGN(SDNode *N, unsigned OpNo);
   SDValue PromoteFloatOp_FP_EXTEND(SDNode *N, unsigned OpNo);
-  SDValue PromoteFloatOp_FP_TO_XINT(SDNode *N, unsigned OpNo);
+  SDValue PromoteFloatOp_UnaryOp(SDNode *N, unsigned OpNo);
   SDValue PromoteFloatOp_FP_TO_XINT_SAT(SDNode *N, unsigned OpNo);
   SDValue PromoteFloatOp_STORE(SDNode *N, unsigned OpNo);
   SDValue PromoteFloatOp_SELECT_CC(SDNode *N, unsigned OpNo);

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

@@ -402,6 +402,8 @@ SDValue VectorLegalizer::LegalizeOp(SDValue Op) {
   case ISD::FCEIL:
   case ISD::FTRUNC:
   case ISD::FRINT:
+  case ISD::LRINT:
+  case ISD::LLRINT:
   case ISD::FNEARBYINT:
   case ISD::FROUND:
   case ISD::FROUNDEVEN:

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

@@ -101,6 +101,8 @@ void DAGTypeLegalizer::ScalarizeVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::FP_TO_SINT:
   case ISD::FP_TO_UINT:
   case ISD::FRINT:
+  case ISD::LRINT:
+  case ISD::LLRINT:
   case ISD::FROUND:
   case ISD::FROUNDEVEN:
   case ISD::FSIN:
@@ -681,6 +683,8 @@ bool DAGTypeLegalizer::ScalarizeVectorOperand(SDNode *N, unsigned OpNo) {
   case ISD::FP_TO_UINT:
   case ISD::SINT_TO_FP:
   case ISD::UINT_TO_FP:
+  case ISD::LRINT:
+  case ISD::LLRINT:
     Res = ScalarizeVecOp_UnaryOp(N);
     break;
   case ISD::STRICT_SINT_TO_FP:
@@ -1097,6 +1101,8 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::VP_FP_TO_UINT:
   case ISD::FRINT:
   case ISD::VP_FRINT:
+  case ISD::LRINT:
+  case ISD::LLRINT:
   case ISD::FROUND:
   case ISD::VP_FROUND:
   case ISD::FROUNDEVEN:
@@ -2974,6 +2980,8 @@ bool DAGTypeLegalizer::SplitVectorOperand(SDNode *N, unsigned OpNo) {
   case ISD::ZERO_EXTEND:
   case ISD::ANY_EXTEND:
   case ISD::FTRUNC:
+  case ISD::LRINT:
+  case ISD::LLRINT:
     Res = SplitVecOp_UnaryOp(N);
     break;
   case ISD::FLDEXP:
@@ -4209,6 +4217,8 @@ void DAGTypeLegalizer::WidenVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::FLOG2:
   case ISD::FNEARBYINT:
   case ISD::FRINT:
+  case ISD::LRINT:
+  case ISD::LLRINT:
   case ISD::FROUND:
   case ISD::FROUNDEVEN:
   case ISD::FSIN:
@@ -5958,7 +5968,11 @@ bool DAGTypeLegalizer::WidenVectorOperand(SDNode *N, unsigned OpNo) {
   case ISD::STRICT_FSETCCS:     Res = WidenVecOp_STRICT_FSETCC(N); break;
   case ISD::VSELECT:            Res = WidenVecOp_VSELECT(N); break;
   case ISD::FLDEXP:
-  case ISD::FCOPYSIGN:          Res = WidenVecOp_UnrollVectorOp(N); break;
+  case ISD::FCOPYSIGN:
+  case ISD::LRINT:
+  case ISD::LLRINT:
+    Res = WidenVecOp_UnrollVectorOp(N);
+    break;
   case ISD::IS_FPCLASS:         Res = WidenVecOp_IS_FPCLASS(N); break;
 
   case ISD::ANY_EXTEND:

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

@@ -5135,6 +5135,8 @@ bool SelectionDAG::isKnownNeverNaN(SDValue Op, bool SNaN, unsigned Depth) const
   case ISD::FROUND:
   case ISD::FROUNDEVEN:
   case ISD::FRINT:
+  case ISD::LRINT:
+  case ISD::LLRINT:
   case ISD::FNEARBYINT:
   case ISD::FLDEXP: {
     if (SNaN)

llvm/lib/CodeGen/TargetLoweringBase.cpp

@@ -873,13 +873,13 @@ void TargetLoweringBase::initActions() {
 
     // These operations default to expand for vector types.
     if (VT.isVector())
-      setOperationAction({ISD::FCOPYSIGN, ISD::SIGN_EXTEND_INREG,
-                          ISD::ANY_EXTEND_VECTOR_INREG,
-                          ISD::SIGN_EXTEND_VECTOR_INREG,
-                          ISD::ZERO_EXTEND_VECTOR_INREG, ISD::SPLAT_VECTOR},
-                         VT, Expand);
+      setOperationAction(
+          {ISD::FCOPYSIGN, ISD::SIGN_EXTEND_INREG, ISD::ANY_EXTEND_VECTOR_INREG,
+           ISD::SIGN_EXTEND_VECTOR_INREG, ISD::ZERO_EXTEND_VECTOR_INREG,
+           ISD::SPLAT_VECTOR, ISD::LRINT, ISD::LLRINT},
+          VT, Expand);
 
-    // Constrained floating-point operations default to expand.
+      // Constrained floating-point operations default to expand.
 #define DAG_INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC, DAGN)               \
     setOperationAction(ISD::STRICT_##DAGN, VT, Expand);
 #include "llvm/IR/ConstrainedOps.def"

llvm/lib/IR/Verifier.cpp

@@ -5669,10 +5669,28 @@ void Verifier::visitIntrinsicCall(Intrinsic::ID ID, CallBase &Call) {
     }
     break;
   }
-  case Intrinsic::lround:
-  case Intrinsic::llround:
   case Intrinsic::lrint:
   case Intrinsic::llrint: {
+    Type *ValTy = Call.getArgOperand(0)->getType();
+    Type *ResultTy = Call.getType();
+    Check(
+        ValTy->isFPOrFPVectorTy() && ResultTy->isIntOrIntVectorTy(),
+        "llvm.lrint, llvm.llrint: argument must be floating-point or vector "
+        "of floating-points, and result must be integer or vector of integers",
+        &Call);
+    Check(ValTy->isVectorTy() == ResultTy->isVectorTy(),
+          "llvm.lrint, llvm.llrint: argument and result disagree on vector use",
+          &Call);
+    if (ValTy->isVectorTy()) {
+      Check(cast<VectorType>(ValTy)->getElementCount() ==
+                cast<VectorType>(ResultTy)->getElementCount(),
+            "llvm.lrint, llvm.llrint: argument must be same length as result",
+            &Call);
+    }
+    break;
+  }
+  case Intrinsic::lround:
+  case Intrinsic::llround: {
     Type *ValTy = Call.getArgOperand(0)->getType();
     Type *ResultTy = Call.getType();
     Check(!ValTy->isVectorTy() && !ResultTy->isVectorTy(),