[LegalizeTypes][VP] Add splitting and widening support for VP_FNEG.

[frasercrmck]

LGTM

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

 //===------- LegalizeVectorTypes.cpp - Legalization of vector types -------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
#endif
   case ISD::FCOS:
   case ISD::FEXP:
   case ISD::FEXP2:
   case ISD::FFLOOR:
   case ISD::FLOG:
   case ISD::FLOG10:
   case ISD::FLOG2:
   case ISD::FNEARBYINT:
-  case ISD::FNEG:
+  case ISD::FNEG: case ISD::VP_FNEG:
   case ISD::FREEZE:
   case ISD::ARITH_FENCE:
   case ISD::FP_EXTEND:
   case ISD::FP_ROUND:
   case ISD::FP_TO_SINT:
   case ISD::FP_TO_UINT:
   case ISD::FRINT:
   case ISD::FROUND:
void DAGTypeLegalizer::SplitVecRes_UnaryOp(SDNode *N, SDValue &Lo,
   // If the input also splits, handle it directly for a compile time speedup.
   // Otherwise split it by hand.
   EVT InVT = N->getOperand(0).getValueType();
   if (getTypeAction(InVT) == TargetLowering::TypeSplitVector)
     GetSplitVector(N->getOperand(0), Lo, Hi);
   else
     std::tie(Lo, Hi) = DAG.SplitVectorOperand(N, 0);
 
-  if (N->getOpcode() == ISD::FP_ROUND) {
-    Lo = DAG.getNode(N->getOpcode(), dl, LoVT, Lo, N->getOperand(1),
-                     N->getFlags());
-    Hi = DAG.getNode(N->getOpcode(), dl, HiVT, Hi, N->getOperand(1),
-                     N->getFlags());
+  const SDNodeFlags Flags = N->getFlags();
+  unsigned Opcode = N->getOpcode();
+  if (N->getNumOperands() <= 2) {
+    if (Opcode == ISD::FP_ROUND) {
+      Lo = DAG.getNode(Opcode, dl, LoVT, Lo, N->getOperand(1), Flags);
+      Hi = DAG.getNode(Opcode, dl, HiVT, Hi, N->getOperand(1), Flags);
     } else {
-    Lo = DAG.getNode(N->getOpcode(), dl, LoVT, Lo, N->getFlags());
-    Hi = DAG.getNode(N->getOpcode(), dl, HiVT, Hi, N->getFlags());
+      Lo = DAG.getNode(Opcode, dl, LoVT, Lo, Flags);
+      Hi = DAG.getNode(Opcode, dl, HiVT, Hi, Flags);
     }
+    return;
+  }
+
+  assert(N->getNumOperands() == 3 && "Unexpected number of operands!");
+  assert(N->isVPOpcode() && "Expected VP opcode");
+
+  SDValue MaskLo, MaskHi;
+  std::tie(MaskLo, MaskHi) = SplitMask(N->getOperand(1));
+
+  SDValue EVLLo, EVLHi;
+  std::tie(EVLLo, EVLHi) =
+      DAG.SplitEVL(N->getOperand(2), N->getValueType(0), dl);
+
+  Lo = DAG.getNode(Opcode, dl, Lo.getValueType(),
+                   {Lo, MaskLo, EVLLo}, Flags);
+  Hi = DAG.getNode(Opcode, dl, Hi.getValueType(),
+                   {Hi, MaskHi, EVLHi}, Flags);
 }
 
 void DAGTypeLegalizer::SplitVecRes_ExtendOp(SDNode *N, SDValue &Lo,
                                             SDValue &Hi) {
   SDLoc dl(N);
   EVT SrcVT = N->getOperand(0).getValueType();
   EVT DestVT = N->getValueType(0);
   EVT LoVT, HiVT;
#include "llvm/IR/ConstrainedOps.def"
   case ISD::ABS:
   case ISD::BITREVERSE:
   case ISD::BSWAP:
   case ISD::CTLZ:
   case ISD::CTLZ_ZERO_UNDEF:
   case ISD::CTPOP:
   case ISD::CTTZ:
   case ISD::CTTZ_ZERO_UNDEF:
-  case ISD::FNEG:
+  case ISD::FNEG: case ISD::VP_FNEG:
   case ISD::FREEZE:
   case ISD::ARITH_FENCE:
   case ISD::FCANONICALIZE:
     Res = WidenVecRes_Unary(N);
     break;
   case ISD::FMA:
   case ISD::FSHL:
   case ISD::FSHR:
SDValue DAGTypeLegalizer::WidenVecRes_POWI(SDNode *N) {
   SDValue ShOp = N->getOperand(1);
   return DAG.getNode(N->getOpcode(), SDLoc(N), WidenVT, InOp, ShOp);
 }
 
 SDValue DAGTypeLegalizer::WidenVecRes_Unary(SDNode *N) {
   // Unary op widening.
   EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   SDValue InOp = GetWidenedVector(N->getOperand(0));
+  if (N->getNumOperands() == 1)
     return DAG.getNode(N->getOpcode(), SDLoc(N), WidenVT, InOp);
+
+  assert(N->getNumOperands() == 3 && "Unexpected number of operands!");
+  assert(N->isVPOpcode() && "Expected VP opcode");
+
+  SDValue Mask =
+      GetWidenedMask(N->getOperand(1), WidenVT.getVectorElementCount());
+  return DAG.getNode(N->getOpcode(), SDLoc(N), WidenVT,
+                     {InOp, Mask, N->getOperand(2)});
 }
 
 SDValue DAGTypeLegalizer::WidenVecRes_InregOp(SDNode *N) {
   EVT WidenVT = TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
   EVT ExtVT = EVT::getVectorVT(*DAG.getContext(),
                                cast<VTSDNode>(N->getOperand(1))->getVT()
                                  .getVectorElementType(),
                                WidenVT.getVectorNumElements());

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vfneg-vp.ll

 ; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
 ; CHECK-NEXT:    ret
   %head = insertelement <8 x i1> poison, i1 true, i32 0
   %m = shufflevector <8 x i1> %head, <8 x i1> poison, <8 x i32> zeroinitializer
   %v = call <8 x double> @llvm.vp.fneg.v8f64(<8 x double> %va, <8 x i1> %m, i32 %evl)
   ret <8 x double> %v
 }
 
+declare <15 x double> @llvm.vp.fneg.v15f64(<15 x double>, <15 x i1>, i32)
+
+define <15 x double> @vfneg_vv_v15f64(<15 x double> %va, <15 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_v15f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vfneg.v v8, v8, v0.t
+; CHECK-NEXT:    ret
+  %v = call <15 x double> @llvm.vp.fneg.v15f64(<15 x double> %va, <15 x i1> %m, i32 %evl)
+  ret <15 x double> %v
+}
+
+define <15 x double> @vfneg_vv_v15f64_unmasked(<15 x double> %va, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_v15f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
+; CHECK-NEXT:    ret
+  %head = insertelement <15 x i1> poison, i1 true, i32 0
+  %m = shufflevector <15 x i1> %head, <15 x i1> poison, <15 x i32> zeroinitializer
+  %v = call <15 x double> @llvm.vp.fneg.v15f64(<15 x double> %va, <15 x i1> %m, i32 %evl)
+  ret <15 x double> %v
+}
+
 declare <16 x double> @llvm.vp.fneg.v16f64(<16 x double>, <16 x i1>, i32)
 
 define <16 x double> @vfneg_vv_v16f64(<16 x double> %va, <16 x i1> %m, i32 zeroext %evl) {
 ; CHECK-LABEL: vfneg_vv_v16f64:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
 ; CHECK-NEXT:    vfneg.v v8, v8, v0.t
 ; CHECK-NEXT:    ret
   %v = call <16 x double> @llvm.vp.fneg.v16f64(<16 x double> %va, <16 x i1> %m, i32 %evl)
   ret <16 x double> %v
 }
 
 define <16 x double> @vfneg_vv_v16f64_unmasked(<16 x double> %va, i32 zeroext %evl) {
 ; CHECK-LABEL: vfneg_vv_v16f64_unmasked:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
 ; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
 ; CHECK-NEXT:    ret
   %head = insertelement <16 x i1> poison, i1 true, i32 0
   %m = shufflevector <16 x i1> %head, <16 x i1> poison, <16 x i32> zeroinitializer
   %v = call <16 x double> @llvm.vp.fneg.v16f64(<16 x double> %va, <16 x i1> %m, i32 %evl)
   ret <16 x double> %v
 }
+
+declare <32 x double> @llvm.vp.fneg.v32f64(<32 x double>, <32 x i1>, i32)
+
+define <32 x double> @vfneg_vv_v32f64(<32 x double> %va, <32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_v32f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vmv1r.v v24, v0
+; CHECK-NEXT:    li a1, 0
+; CHECK-NEXT:    vsetivli zero, 2, e8, mf4, ta, mu
+; CHECK-NEXT:    addi a2, a0, -16
+; CHECK-NEXT:    vslidedown.vi v0, v0, 2
+; CHECK-NEXT:    bltu a0, a2, .LBB26_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a1, a2
+; CHECK-NEXT:  .LBB26_2:
+; CHECK-NEXT:    vsetvli zero, a1, e64, m8, ta, mu
+; CHECK-NEXT:    li a1, 16
+; CHECK-NEXT:    vfneg.v v16, v16, v0.t
+; CHECK-NEXT:    bltu a0, a1, .LBB26_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    li a0, 16
+; CHECK-NEXT:  .LBB26_4:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vmv1r.v v0, v24
+; CHECK-NEXT:    vfneg.v v8, v8, v0.t
+; CHECK-NEXT:    ret
+  %v = call <32 x double> @llvm.vp.fneg.v32f64(<32 x double> %va, <32 x i1> %m, i32 %evl)
+  ret <32 x double> %v
+}
+
+define <32 x double> @vfneg_vv_v32f64_unmasked(<32 x double> %va, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_v32f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, a0, -16
+; CHECK-NEXT:    li a2, 0
+; CHECK-NEXT:    bltu a0, a1, .LBB27_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a2, a1
+; CHECK-NEXT:  .LBB27_2:
+; CHECK-NEXT:    vsetvli zero, a2, e64, m8, ta, mu
+; CHECK-NEXT:    li a1, 16
+; CHECK-NEXT:    vfsgnjn.vv v16, v16, v16
+; CHECK-NEXT:    bltu a0, a1, .LBB27_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    li a0, 16
+; CHECK-NEXT:  .LBB27_4:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
+; CHECK-NEXT:    ret
+  %head = insertelement <32 x i1> poison, i1 true, i32 0
+  %m = shufflevector <32 x i1> %head, <32 x i1> poison, <32 x i32> zeroinitializer
+  %v = call <32 x double> @llvm.vp.fneg.v32f64(<32 x double> %va, <32 x i1> %m, i32 %evl)
+  ret <32 x double> %v
+}

llvm/test/CodeGen/RISCV/rvv/vfneg-vp.ll

 ; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
 ; CHECK-NEXT:    ret
   %head = insertelement <vscale x 4 x i1> poison, i1 true, i32 0
   %m = shufflevector <vscale x 4 x i1> %head, <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer
   %v = call <vscale x 4 x double> @llvm.vp.fneg.nxv4f64(<vscale x 4 x double> %va, <vscale x 4 x i1> %m, i32 %evl)
   ret <vscale x 4 x double> %v
 }
 
+declare <vscale x 7 x double> @llvm.vp.fneg.nxv7f64(<vscale x 7 x double>, <vscale x 7 x i1>, i32)
+
+define <vscale x 7 x double> @vfneg_vv_nxv7f64(<vscale x 7 x double> %va, <vscale x 7 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_nxv7f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vfneg.v v8, v8, v0.t
+; CHECK-NEXT:    ret
+  %v = call <vscale x 7 x double> @llvm.vp.fneg.nxv7f64(<vscale x 7 x double> %va, <vscale x 7 x i1> %m, i32 %evl)
+  ret <vscale x 7 x double> %v
+}
+
+define <vscale x 7 x double> @vfneg_vv_nxv7f64_unmasked(<vscale x 7 x double> %va, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_nxv7f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
+; CHECK-NEXT:    ret
+  %head = insertelement <vscale x 7 x i1> poison, i1 true, i32 0
+  %m = shufflevector <vscale x 7 x i1> %head, <vscale x 7 x i1> poison, <vscale x 7 x i32> zeroinitializer
+  %v = call <vscale x 7 x double> @llvm.vp.fneg.nxv7f64(<vscale x 7 x double> %va, <vscale x 7 x i1> %m, i32 %evl)
+  ret <vscale x 7 x double> %v
+}
+
 declare <vscale x 8 x double> @llvm.vp.fneg.nxv8f64(<vscale x 8 x double>, <vscale x 8 x i1>, i32)
 
 define <vscale x 8 x double> @vfneg_vv_nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x i1> %m, i32 zeroext %evl) {
 ; CHECK-LABEL: vfneg_vv_nxv8f64:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
 ; CHECK-NEXT:    vfneg.v v8, v8, v0.t
 ; CHECK-NEXT:    ret
   %v = call <vscale x 8 x double> @llvm.vp.fneg.nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x i1> %m, i32 %evl)
   ret <vscale x 8 x double> %v
 }
 
 define <vscale x 8 x double> @vfneg_vv_nxv8f64_unmasked(<vscale x 8 x double> %va, i32 zeroext %evl) {
 ; CHECK-LABEL: vfneg_vv_nxv8f64_unmasked:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
 ; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
 ; CHECK-NEXT:    ret
   %head = insertelement <vscale x 8 x i1> poison, i1 true, i32 0
   %m = shufflevector <vscale x 8 x i1> %head, <vscale x 8 x i1> poison, <vscale x 8 x i32> zeroinitializer
   %v = call <vscale x 8 x double> @llvm.vp.fneg.nxv8f64(<vscale x 8 x double> %va, <vscale x 8 x i1> %m, i32 %evl)
   ret <vscale x 8 x double> %v
 }
+
+; Test splitting.
+declare <vscale x 16 x double> @llvm.vp.fneg.nxv16f64(<vscale x 16 x double>, <vscale x 16 x i1>, i32)
+
+define <vscale x 16 x double> @vfneg_vv_nxv16f64(<vscale x 16 x double> %va, <vscale x 16 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_nxv16f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vmv1r.v v24, v0
+; CHECK-NEXT:    li a2, 0
+; CHECK-NEXT:    csrr a1, vlenb
+; CHECK-NEXT:    srli a4, a1, 3
+; CHECK-NEXT:    vsetvli a3, zero, e8, mf4, ta, mu
+; CHECK-NEXT:    sub a3, a0, a1
+; CHECK-NEXT:    vslidedown.vx v0, v0, a4
+; CHECK-NEXT:    bltu a0, a3, .LBB32_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a2, a3
+; CHECK-NEXT:  .LBB32_2:
+; CHECK-NEXT:    vsetvli zero, a2, e64, m8, ta, mu
+; CHECK-NEXT:    vfneg.v v16, v16, v0.t
+; CHECK-NEXT:    bltu a0, a1, .LBB32_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    mv a0, a1
+; CHECK-NEXT:  .LBB32_4:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, ta, mu
+; CHECK-NEXT:    vmv1r.v v0, v24
+; CHECK-NEXT:    vfneg.v v8, v8, v0.t
+; CHECK-NEXT:    ret
+  %v = call <vscale x 16 x double> @llvm.vp.fneg.nxv16f64(<vscale x 16 x double> %va, <vscale x 16 x i1> %m, i32 %evl)
+  ret <vscale x 16 x double> %v
+}
+
+define <vscale x 16 x double> @vfneg_vv_nxv16f64_unmasked(<vscale x 16 x double> %va, i32 zeroext %evl) {
+; CHECK-LABEL: vfneg_vv_nxv16f64_unmasked:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    csrr a1, vlenb
+; CHECK-NEXT:    mv a2, a0
+; CHECK-NEXT:    bltu a0, a1, .LBB33_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a2, a1
+; CHECK-NEXT:  .LBB33_2:
+; CHECK-NEXT:    li a3, 0
+; CHECK-NEXT:    vsetvli zero, a2, e64, m8, ta, mu
+; CHECK-NEXT:    sub a1, a0, a1
+; CHECK-NEXT:    vfsgnjn.vv v8, v8, v8
+; CHECK-NEXT:    bltu a0, a1, .LBB33_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    mv a3, a1
+; CHECK-NEXT:  .LBB33_4:
+; CHECK-NEXT:    vsetvli zero, a3, e64, m8, ta, mu
+; CHECK-NEXT:    vfsgnjn.vv v16, v16, v16
+; CHECK-NEXT:    ret
+  %head = insertelement <vscale x 16 x i1> poison, i1 true, i32 0
+  %m = shufflevector <vscale x 16 x i1> %head, <vscale x 16 x i1> poison, <vscale x 16 x i32> zeroinitializer
+  %v = call <vscale x 16 x double> @llvm.vp.fneg.nxv16f64(<vscale x 16 x double> %va, <vscale x 16 x i1> %m, i32 %evl)
+  ret <vscale x 16 x double> %v
+}