[RISCV] Add basic cost model for vp float rounding instructions.

[fakepaper56]

Rebase and ping.

[reames]

LGTM

[fakepaper56]

Use mapped isd opcode on isOperationCustom for each intrinsic id.

[craig.topper]

LGTM

llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp

static const CostTblEntry VectorIntrinsicCostTable[]{
     {Intrinsic::ctpop, MVT::v8i64, 21},
     {Intrinsic::ctpop, MVT::v16i64, 21},
     {Intrinsic::ctpop, MVT::nxv1i64, 21},
     {Intrinsic::ctpop, MVT::nxv2i64, 21},
     {Intrinsic::ctpop, MVT::nxv4i64, 21},
     {Intrinsic::ctpop, MVT::nxv8i64, 21},
 };
 
+static unsigned getISDForVPIntrinsicID(Intrinsic::ID ID) {
+  switch (ID) {
+#define HELPER_MAP_VPID_TO_VPSD(VPID, VPSD)                                    \
+  case Intrinsic::VPID:                                                        \
+    return ISD::VPSD;
+#include "llvm/IR/VPIntrinsics.def"
+#undef HELPER_MAP_VPID_TO_VPSD
+  }
+  return ISD::DELETED_NODE;
+}
+
 InstructionCost
 RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
                                     TTI::TargetCostKind CostKind) {
   auto *RetTy = ICA.getReturnType();
   switch (ICA.getID()) {
   case Intrinsic::ceil:
   case Intrinsic::floor:
   case Intrinsic::trunc:
RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
   case Intrinsic::vp_nearbyint: {
     // More one read and one write for fflags than vp_rint.
     unsigned Cost = 7;
     auto LT = getTypeLegalizationCost(RetTy);
     if (TLI->isOperationCustom(ISD::VP_FRINT, LT.second))
       return Cost * LT.first;
     break;
   }
+  case Intrinsic::vp_ceil:
+  case Intrinsic::vp_floor:
+  case Intrinsic::vp_round:
+  case Intrinsic::vp_roundeven:
+  case Intrinsic::vp_roundtozero: {
+    // Rounding with static rounding mode needs two more instructions to
+    // swap/write FRM than vp_rint.
+    unsigned Cost = 7;
+    auto LT = getTypeLegalizationCost(RetTy);
+    unsigned VPISD = getISDForVPIntrinsicID(ICA.getID());
+    if (TLI->isOperationCustom(VPISD, LT.second))
+      return Cost * LT.first;
+    break;
+  }
   }
 
   if (ST->hasVInstructions() && RetTy->isVectorTy()) {
     auto LT = getTypeLegalizationCost(RetTy);
     if (const auto *Entry = CostTableLookup(VectorIntrinsicCostTable,
                                             ICA.getID(), LT.second))
       return LT.first * Entry->Cost;
   }

llvm/test/Analysis/CostModel/RISCV/fround.ll

;
   call <16 x double> @llvm.roundeven.v16f64(<16 x double> undef)
   call <vscale x 1 x double> @llvm.roundeven.nvx1f64(<vscale x 1 x double> undef)
   call <vscale x 2 x double> @llvm.roundeven.nvx2f64(<vscale x 2 x double> undef)
   call <vscale x 4 x double> @llvm.roundeven.nvx5f64(<vscale x 4 x double> undef)
   call <vscale x 8 x double> @llvm.roundeven.nvx8f64(<vscale x 8 x double> undef)
   ret void
 }
 
+define void @vp_ceil() {
+; CHECK-LABEL: 'vp_ceil'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %1 = call <2 x float> @llvm.vp.ceil.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %2 = call <4 x float> @llvm.vp.ceil.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %3 = call <8 x float> @llvm.vp.ceil.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %4 = call <16 x float> @llvm.vp.ceil.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %5 = call <vscale x 1 x float> @llvm.vp.ceil.nxv1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %6 = call <vscale x 2 x float> @llvm.vp.ceil.nxv2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %7 = call <vscale x 4 x float> @llvm.vp.ceil.nxv4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %8 = call <vscale x 8 x float> @llvm.vp.ceil.nxv8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %9 = call <vscale x 16 x float> @llvm.vp.ceil.nxv16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %10 = call <2 x double> @llvm.vp.ceil.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %11 = call <4 x double> @llvm.vp.ceil.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %12 = call <8 x double> @llvm.vp.ceil.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %13 = call <16 x double> @llvm.vp.ceil.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %14 = call <vscale x 1 x double> @llvm.vp.ceil.nxv1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %15 = call <vscale x 2 x double> @llvm.vp.ceil.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %16 = call <vscale x 4 x double> @llvm.vp.ceil.nxv4f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %17 = call <vscale x 8 x double> @llvm.vp.ceil.nxv8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  call <2 x float> @llvm.vp.ceil.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+  call <4 x float> @llvm.vp.ceil.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+  call <8 x float> @llvm.vp.ceil.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+  call <16 x float> @llvm.vp.ceil.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x float> @llvm.vp.ceil.nvx1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x float> @llvm.vp.ceil.nvx2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x float> @llvm.vp.ceil.nvx4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x float> @llvm.vp.ceil.nvx8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+  call <vscale x 16 x float> @llvm.vp.ceil.nvx16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+  call <2 x double> @llvm.vp.ceil.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+  call <4 x double> @llvm.vp.ceil.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+  call <8 x double> @llvm.vp.ceil.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+  call <16 x double> @llvm.vp.ceil.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x double> @llvm.vp.ceil.nvx1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x double> @llvm.vp.ceil.nvx2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x double> @llvm.vp.ceil.nvx5f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x double> @llvm.vp.ceil.nvx8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+  ret void
+}
+
+define void @vp_floor() {
+; CHECK-LABEL: 'vp_floor'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %1 = call <2 x float> @llvm.vp.floor.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %2 = call <4 x float> @llvm.vp.floor.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %3 = call <8 x float> @llvm.vp.floor.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %4 = call <16 x float> @llvm.vp.floor.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %5 = call <vscale x 1 x float> @llvm.vp.floor.nxv1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %6 = call <vscale x 2 x float> @llvm.vp.floor.nxv2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %7 = call <vscale x 4 x float> @llvm.vp.floor.nxv4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %8 = call <vscale x 8 x float> @llvm.vp.floor.nxv8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %9 = call <vscale x 16 x float> @llvm.vp.floor.nxv16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %10 = call <2 x double> @llvm.vp.floor.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %11 = call <4 x double> @llvm.vp.floor.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %12 = call <8 x double> @llvm.vp.floor.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %13 = call <16 x double> @llvm.vp.floor.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %14 = call <vscale x 1 x double> @llvm.vp.floor.nxv1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %15 = call <vscale x 2 x double> @llvm.vp.floor.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %16 = call <vscale x 4 x double> @llvm.vp.floor.nxv4f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %17 = call <vscale x 8 x double> @llvm.vp.floor.nxv8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  call <2 x float> @llvm.vp.floor.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+  call <4 x float> @llvm.vp.floor.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+  call <8 x float> @llvm.vp.floor.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+  call <16 x float> @llvm.vp.floor.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x float> @llvm.vp.floor.nvx1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x float> @llvm.vp.floor.nvx2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x float> @llvm.vp.floor.nvx4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x float> @llvm.vp.floor.nvx8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+  call <vscale x 16 x float> @llvm.vp.floor.nvx16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+  call <2 x double> @llvm.vp.floor.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+  call <4 x double> @llvm.vp.floor.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+  call <8 x double> @llvm.vp.floor.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+  call <16 x double> @llvm.vp.floor.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x double> @llvm.vp.floor.nvx1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x double> @llvm.vp.floor.nvx2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x double> @llvm.vp.floor.nvx5f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x double> @llvm.vp.floor.nvx8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+  ret void
+}
+
+define void @vp_round() {
+; CHECK-LABEL: 'vp_round'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %1 = call <2 x float> @llvm.vp.round.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %2 = call <4 x float> @llvm.vp.round.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %3 = call <8 x float> @llvm.vp.round.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %4 = call <16 x float> @llvm.vp.round.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %5 = call <vscale x 1 x float> @llvm.vp.round.nxv1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %6 = call <vscale x 2 x float> @llvm.vp.round.nxv2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %7 = call <vscale x 4 x float> @llvm.vp.round.nxv4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %8 = call <vscale x 8 x float> @llvm.vp.round.nxv8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %9 = call <vscale x 16 x float> @llvm.vp.round.nxv16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %10 = call <2 x double> @llvm.vp.round.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %11 = call <4 x double> @llvm.vp.round.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %12 = call <8 x double> @llvm.vp.round.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %13 = call <16 x double> @llvm.vp.round.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %14 = call <vscale x 1 x double> @llvm.vp.round.nxv1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %15 = call <vscale x 2 x double> @llvm.vp.round.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %16 = call <vscale x 4 x double> @llvm.vp.round.nxv4f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %17 = call <vscale x 8 x double> @llvm.vp.round.nxv8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  call <2 x float> @llvm.vp.round.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+  call <4 x float> @llvm.vp.round.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+  call <8 x float> @llvm.vp.round.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+  call <16 x float> @llvm.vp.round.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x float> @llvm.vp.round.nvx1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x float> @llvm.vp.round.nvx2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x float> @llvm.vp.round.nvx4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x float> @llvm.vp.round.nvx8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+  call <vscale x 16 x float> @llvm.vp.round.nvx16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+  call <2 x double> @llvm.vp.round.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+  call <4 x double> @llvm.vp.round.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+  call <8 x double> @llvm.vp.round.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+  call <16 x double> @llvm.vp.round.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x double> @llvm.vp.round.nvx1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x double> @llvm.vp.round.nvx2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x double> @llvm.vp.round.nvx5f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x double> @llvm.vp.round.nvx8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+  ret void
+}
+
+define void @vp_roundeven() {
+; CHECK-LABEL: 'vp_roundeven'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %1 = call <2 x float> @llvm.vp.roundeven.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %2 = call <4 x float> @llvm.vp.roundeven.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %3 = call <8 x float> @llvm.vp.roundeven.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %4 = call <16 x float> @llvm.vp.roundeven.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %5 = call <vscale x 1 x float> @llvm.vp.roundeven.nxv1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %6 = call <vscale x 2 x float> @llvm.vp.roundeven.nxv2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %7 = call <vscale x 4 x float> @llvm.vp.roundeven.nxv4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %8 = call <vscale x 8 x float> @llvm.vp.roundeven.nxv8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %9 = call <vscale x 16 x float> @llvm.vp.roundeven.nxv16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %10 = call <2 x double> @llvm.vp.roundeven.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %11 = call <4 x double> @llvm.vp.roundeven.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %12 = call <8 x double> @llvm.vp.roundeven.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %13 = call <16 x double> @llvm.vp.roundeven.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %14 = call <vscale x 1 x double> @llvm.vp.roundeven.nxv1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %15 = call <vscale x 2 x double> @llvm.vp.roundeven.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %16 = call <vscale x 4 x double> @llvm.vp.roundeven.nxv4f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %17 = call <vscale x 8 x double> @llvm.vp.roundeven.nxv8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  call <2 x float> @llvm.vp.roundeven.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+  call <4 x float> @llvm.vp.roundeven.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+  call <8 x float> @llvm.vp.roundeven.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+  call <16 x float> @llvm.vp.roundeven.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x float> @llvm.vp.roundeven.nvx1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x float> @llvm.vp.roundeven.nvx2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x float> @llvm.vp.roundeven.nvx4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x float> @llvm.vp.roundeven.nvx8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+  call <vscale x 16 x float> @llvm.vp.roundeven.nvx16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+  call <2 x double> @llvm.vp.roundeven.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+  call <4 x double> @llvm.vp.roundeven.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+  call <8 x double> @llvm.vp.roundeven.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+  call <16 x double> @llvm.vp.roundeven.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x double> @llvm.vp.roundeven.nvx1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x double> @llvm.vp.roundeven.nvx2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x double> @llvm.vp.roundeven.nvx5f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x double> @llvm.vp.roundeven.nvx8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+  ret void
+}
+
+define void @vp_roundtozero() {
+; CHECK-LABEL: 'vp_roundtozero'
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %1 = call <2 x float> @llvm.vp.roundtozero.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %2 = call <4 x float> @llvm.vp.roundtozero.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %3 = call <8 x float> @llvm.vp.roundtozero.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %4 = call <16 x float> @llvm.vp.roundtozero.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %5 = call <vscale x 1 x float> @llvm.vp.roundtozero.nxv1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %6 = call <vscale x 2 x float> @llvm.vp.roundtozero.nxv2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %7 = call <vscale x 4 x float> @llvm.vp.roundtozero.nxv4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %8 = call <vscale x 8 x float> @llvm.vp.roundtozero.nxv8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %9 = call <vscale x 16 x float> @llvm.vp.roundtozero.nxv16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %10 = call <2 x double> @llvm.vp.roundtozero.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %11 = call <4 x double> @llvm.vp.roundtozero.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %12 = call <8 x double> @llvm.vp.roundtozero.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %13 = call <16 x double> @llvm.vp.roundtozero.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %14 = call <vscale x 1 x double> @llvm.vp.roundtozero.nxv1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %15 = call <vscale x 2 x double> @llvm.vp.roundtozero.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %16 = call <vscale x 4 x double> @llvm.vp.roundtozero.nxv4f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %17 = call <vscale x 8 x double> @llvm.vp.roundtozero.nxv8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  call <2 x float> @llvm.vp.roundtozero.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
+  call <4 x float> @llvm.vp.roundtozero.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
+  call <8 x float> @llvm.vp.roundtozero.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
+  call <16 x float> @llvm.vp.roundtozero.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x float> @llvm.vp.roundtozero.nvx1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x float> @llvm.vp.roundtozero.nvx2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x float> @llvm.vp.roundtozero.nvx4f32(<vscale x 4 x float> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x float> @llvm.vp.roundtozero.nvx8f32(<vscale x 8 x float> undef, <vscale x 8 x i1> undef, i32 undef)
+  call <vscale x 16 x float> @llvm.vp.roundtozero.nvx16f32(<vscale x 16 x float> undef, <vscale x 16 x i1> undef, i32 undef)
+  call <2 x double> @llvm.vp.roundtozero.v2f64(<2 x double> undef, <2 x i1> undef, i32 undef)
+  call <4 x double> @llvm.vp.roundtozero.v4f64(<4 x double> undef, <4 x i1> undef, i32 undef)
+  call <8 x double> @llvm.vp.roundtozero.v8f64(<8 x double> undef, <8 x i1> undef, i32 undef)
+  call <16 x double> @llvm.vp.roundtozero.v16f64(<16 x double> undef, <16 x i1> undef, i32 undef)
+  call <vscale x 1 x double> @llvm.vp.roundtozero.nvx1f64(<vscale x 1 x double> undef, <vscale x 1 x i1> undef, i32 undef)
+  call <vscale x 2 x double> @llvm.vp.roundtozero.nvx2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> undef, i32 undef)
+  call <vscale x 4 x double> @llvm.vp.roundtozero.nvx5f64(<vscale x 4 x double> undef, <vscale x 4 x i1> undef, i32 undef)
+  call <vscale x 8 x double> @llvm.vp.roundtozero.nvx8f64(<vscale x 8 x double> undef, <vscale x 8 x i1> undef, i32 undef)
+  ret void
+}
+
 define void @vp_rint() {
 ; CHECK-LABEL: 'vp_rint'
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %1 = call <2 x float> @llvm.vp.rint.v2f32(<2 x float> undef, <2 x i1> undef, i32 undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %2 = call <4 x float> @llvm.vp.rint.v4f32(<4 x float> undef, <4 x i1> undef, i32 undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %3 = call <8 x float> @llvm.vp.rint.v8f32(<8 x float> undef, <8 x i1> undef, i32 undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %4 = call <16 x float> @llvm.vp.rint.v16f32(<16 x float> undef, <16 x i1> undef, i32 undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %5 = call <vscale x 1 x float> @llvm.vp.rint.nxv1f32(<vscale x 1 x float> undef, <vscale x 1 x i1> undef, i32 undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %6 = call <vscale x 2 x float> @llvm.vp.rint.nxv2f32(<vscale x 2 x float> undef, <vscale x 2 x i1> undef, i32 undef)

 declare <4 x double> @llvm.roundeven.v4f64(<4 x double>)
 declare <8 x double> @llvm.roundeven.v8f64(<8 x double>)
 declare <16 x double> @llvm.roundeven.v16f64(<16 x double>)
 declare <vscale x 1 x double> @llvm.roundeven.nvx1f64(<vscale x 1 x double>)
 declare <vscale x 2 x double> @llvm.roundeven.nvx2f64(<vscale x 2 x double>)
 declare <vscale x 4 x double> @llvm.roundeven.nvx5f64(<vscale x 4 x double>)
 declare <vscale x 8 x double> @llvm.roundeven.nvx8f64(<vscale x 8 x double>)
 
+declare <2 x float> @llvm.vp.ceil.v2f32(<2 x float>, <2 x i1>, i32)
+declare <4 x float> @llvm.vp.ceil.v4f32(<4 x float>, <4 x i1>, i32)
+declare <8 x float> @llvm.vp.ceil.v8f32(<8 x float>, <8 x i1>, i32)
+declare <16 x float> @llvm.vp.ceil.v16f32(<16 x float>, <16 x i1>, i32)
+declare <vscale x 1 x float> @llvm.vp.ceil.nvx1f32(<vscale x 1 x float>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x float> @llvm.vp.ceil.nvx2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x float> @llvm.vp.ceil.nvx4f32(<vscale x 4 x float>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x float> @llvm.vp.ceil.nvx8f32(<vscale x 8 x float>, <vscale x 8 x i1>, i32)
+declare <vscale x 16 x float> @llvm.vp.ceil.nvx16f32(<vscale x 16 x float>, <vscale x 16 x i1>, i32)
+declare double @llvm.vp.ceil.f64(double)
+declare <2 x double> @llvm.vp.ceil.v2f64(<2 x double>, <2 x i1>, i32)
+declare <4 x double> @llvm.vp.ceil.v4f64(<4 x double>, <4 x i1>, i32)
+declare <8 x double> @llvm.vp.ceil.v8f64(<8 x double>, <8 x i1>, i32)
+declare <16 x double> @llvm.vp.ceil.v16f64(<16 x double>, <16 x i1>, i32)
+declare <vscale x 1 x double> @llvm.vp.ceil.nvx1f64(<vscale x 1 x double>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x double> @llvm.vp.ceil.nvx2f64(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x double> @llvm.vp.ceil.nvx5f64(<vscale x 4 x double>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x double> @llvm.vp.ceil.nvx8f64(<vscale x 8 x double>, <vscale x 8 x i1>, i32)
+
+declare <2 x float> @llvm.vp.floor.v2f32(<2 x float>, <2 x i1>, i32)
+declare <4 x float> @llvm.vp.floor.v4f32(<4 x float>, <4 x i1>, i32)
+declare <8 x float> @llvm.vp.floor.v8f32(<8 x float>, <8 x i1>, i32)
+declare <16 x float> @llvm.vp.floor.v16f32(<16 x float>, <16 x i1>, i32)
+declare <vscale x 1 x float> @llvm.vp.floor.nvx1f32(<vscale x 1 x float>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x float> @llvm.vp.floor.nvx2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x float> @llvm.vp.floor.nvx4f32(<vscale x 4 x float>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x float> @llvm.vp.floor.nvx8f32(<vscale x 8 x float>, <vscale x 8 x i1>, i32)
+declare <vscale x 16 x float> @llvm.vp.floor.nvx16f32(<vscale x 16 x float>, <vscale x 16 x i1>, i32)
+declare double @llvm.vp.floor.f64(double)
+declare <2 x double> @llvm.vp.floor.v2f64(<2 x double>, <2 x i1>, i32)
+declare <4 x double> @llvm.vp.floor.v4f64(<4 x double>, <4 x i1>, i32)
+declare <8 x double> @llvm.vp.floor.v8f64(<8 x double>, <8 x i1>, i32)
+declare <16 x double> @llvm.vp.floor.v16f64(<16 x double>, <16 x i1>, i32)
+declare <vscale x 1 x double> @llvm.vp.floor.nvx1f64(<vscale x 1 x double>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x double> @llvm.vp.floor.nvx2f64(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x double> @llvm.vp.floor.nvx5f64(<vscale x 4 x double>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x double> @llvm.vp.floor.nvx8f64(<vscale x 8 x double>, <vscale x 8 x i1>, i32)
+
+declare <2 x float> @llvm.vp.round.v2f32(<2 x float>, <2 x i1>, i32)
+declare <4 x float> @llvm.vp.round.v4f32(<4 x float>, <4 x i1>, i32)
+declare <8 x float> @llvm.vp.round.v8f32(<8 x float>, <8 x i1>, i32)
+declare <16 x float> @llvm.vp.round.v16f32(<16 x float>, <16 x i1>, i32)
+declare <vscale x 1 x float> @llvm.vp.round.nvx1f32(<vscale x 1 x float>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x float> @llvm.vp.round.nvx2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x float> @llvm.vp.round.nvx4f32(<vscale x 4 x float>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x float> @llvm.vp.round.nvx8f32(<vscale x 8 x float>, <vscale x 8 x i1>, i32)
+declare <vscale x 16 x float> @llvm.vp.round.nvx16f32(<vscale x 16 x float>, <vscale x 16 x i1>, i32)
+declare double @llvm.vp.round.f64(double)
+declare <2 x double> @llvm.vp.round.v2f64(<2 x double>, <2 x i1>, i32)
+declare <4 x double> @llvm.vp.round.v4f64(<4 x double>, <4 x i1>, i32)
+declare <8 x double> @llvm.vp.round.v8f64(<8 x double>, <8 x i1>, i32)
+declare <16 x double> @llvm.vp.round.v16f64(<16 x double>, <16 x i1>, i32)
+declare <vscale x 1 x double> @llvm.vp.round.nvx1f64(<vscale x 1 x double>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x double> @llvm.vp.round.nvx2f64(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x double> @llvm.vp.round.nvx5f64(<vscale x 4 x double>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x double> @llvm.vp.round.nvx8f64(<vscale x 8 x double>, <vscale x 8 x i1>, i32)
+
+declare <2 x float> @llvm.vp.roundeven.v2f32(<2 x float>, <2 x i1>, i32)
+declare <4 x float> @llvm.vp.roundeven.v4f32(<4 x float>, <4 x i1>, i32)
+declare <8 x float> @llvm.vp.roundeven.v8f32(<8 x float>, <8 x i1>, i32)
+declare <16 x float> @llvm.vp.roundeven.v16f32(<16 x float>, <16 x i1>, i32)
+declare <vscale x 1 x float> @llvm.vp.roundeven.nvx1f32(<vscale x 1 x float>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x float> @llvm.vp.roundeven.nvx2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x float> @llvm.vp.roundeven.nvx4f32(<vscale x 4 x float>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x float> @llvm.vp.roundeven.nvx8f32(<vscale x 8 x float>, <vscale x 8 x i1>, i32)
+declare <vscale x 16 x float> @llvm.vp.roundeven.nvx16f32(<vscale x 16 x float>, <vscale x 16 x i1>, i32)
+declare double @llvm.vp.roundeven.f64(double)
+declare <2 x double> @llvm.vp.roundeven.v2f64(<2 x double>, <2 x i1>, i32)
+declare <4 x double> @llvm.vp.roundeven.v4f64(<4 x double>, <4 x i1>, i32)
+declare <8 x double> @llvm.vp.roundeven.v8f64(<8 x double>, <8 x i1>, i32)
+declare <16 x double> @llvm.vp.roundeven.v16f64(<16 x double>, <16 x i1>, i32)
+declare <vscale x 1 x double> @llvm.vp.roundeven.nvx1f64(<vscale x 1 x double>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x double> @llvm.vp.roundeven.nvx2f64(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x double> @llvm.vp.roundeven.nvx5f64(<vscale x 4 x double>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x double> @llvm.vp.roundeven.nvx8f64(<vscale x 8 x double>, <vscale x 8 x i1>, i32)
+
+declare <2 x float> @llvm.vp.roundtozero.v2f32(<2 x float>, <2 x i1>, i32)
+declare <4 x float> @llvm.vp.roundtozero.v4f32(<4 x float>, <4 x i1>, i32)
+declare <8 x float> @llvm.vp.roundtozero.v8f32(<8 x float>, <8 x i1>, i32)
+declare <16 x float> @llvm.vp.roundtozero.v16f32(<16 x float>, <16 x i1>, i32)
+declare <vscale x 1 x float> @llvm.vp.roundtozero.nvx1f32(<vscale x 1 x float>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x float> @llvm.vp.roundtozero.nvx2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x float> @llvm.vp.roundtozero.nvx4f32(<vscale x 4 x float>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x float> @llvm.vp.roundtozero.nvx8f32(<vscale x 8 x float>, <vscale x 8 x i1>, i32)
+declare <vscale x 16 x float> @llvm.vp.roundtozero.nvx16f32(<vscale x 16 x float>, <vscale x 16 x i1>, i32)
+declare double @llvm.vp.roundtozero.f64(double)
+declare <2 x double> @llvm.vp.roundtozero.v2f64(<2 x double>, <2 x i1>, i32)
+declare <4 x double> @llvm.vp.roundtozero.v4f64(<4 x double>, <4 x i1>, i32)
+declare <8 x double> @llvm.vp.roundtozero.v8f64(<8 x double>, <8 x i1>, i32)
+declare <16 x double> @llvm.vp.roundtozero.v16f64(<16 x double>, <16 x i1>, i32)
+declare <vscale x 1 x double> @llvm.vp.roundtozero.nvx1f64(<vscale x 1 x double>, <vscale x 1 x i1>, i32)
+declare <vscale x 2 x double> @llvm.vp.roundtozero.nvx2f64(<vscale x 2 x double>, <vscale x 2 x i1>, i32)
+declare <vscale x 4 x double> @llvm.vp.roundtozero.nvx5f64(<vscale x 4 x double>, <vscale x 4 x i1>, i32)
+declare <vscale x 8 x double> @llvm.vp.roundtozero.nvx8f64(<vscale x 8 x double>, <vscale x 8 x i1>, i32)
+
 declare <2 x float> @llvm.vp.rint.v2f32(<2 x float>, <2 x i1>, i32)
 declare <4 x float> @llvm.vp.rint.v4f32(<4 x float>, <4 x i1>, i32)
 declare <8 x float> @llvm.vp.rint.v8f32(<8 x float>, <8 x i1>, i32)
 declare <16 x float> @llvm.vp.rint.v16f32(<16 x float>, <16 x i1>, i32)
 declare <vscale x 1 x float> @llvm.vp.rint.nvx1f32(<vscale x 1 x float>, <vscale x 1 x i1>, i32)
 declare <vscale x 2 x float> @llvm.vp.rint.nvx2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
 declare <vscale x 4 x float> @llvm.vp.rint.nvx4f32(<vscale x 4 x float>, <vscale x 4 x i1>, i32)
 declare <vscale x 8 x float> @llvm.vp.rint.nvx8f32(<vscale x 8 x float>, <vscale x 8 x i1>, i32)