[SelectionDAG][VP] Add splitting support for VP_MERGE

[frasercrmck]

Summary

This patch adds splitting support for ISD::VP_MERGE, which splits
identically to VP_SELECT and similarly to other select-like nodes.

[craig.topper]

LGTM

llvm/lib/CodeGen/SelectionDAG/LegalizeTypesGeneric.cpp

else if (Cond.getOpcode() == ISD::SETCC) {
           getSetCCResultType(CondLHSVT) == Cond.getValueType())
         std::tie(CL, CH) = DAG.SplitVector(Cond, dl);
       else
         SplitVecRes_SETCC(Cond.getNode(), CL, CH);
     } else
       std::tie(CL, CH) = DAG.SplitVector(Cond, dl);
   }
 
-  if (Opcode != ISD::VP_SELECT) {
+  if (Opcode != ISD::VP_SELECT && Opcode != ISD::VP_MERGE) {
     Lo = DAG.getNode(Opcode, dl, LL.getValueType(), CL, LL, RL);
     Hi = DAG.getNode(Opcode, dl, LH.getValueType(), CH, LH, RH);
     return;
   }
 
   SDValue EVLLo, EVLHi;
   std::tie(EVLLo, EVLHi) =
       DAG.SplitEVL(N->getOperand(3), N->getValueType(0), dl);

llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

#ifndef NDEBUG
     dbgs() << "\n";
 #endif
     report_fatal_error("Do not know how to split the result of this "
                        "operator!\n");
 
   case ISD::MERGE_VALUES: SplitRes_MERGE_VALUES(N, ResNo, Lo, Hi); break;
   case ISD::VSELECT:
   case ISD::SELECT:
+  case ISD::VP_MERGE:
   case ISD::VP_SELECT:    SplitRes_Select(N, Lo, Hi); break;
   case ISD::SELECT_CC:    SplitRes_SELECT_CC(N, Lo, Hi); break;
   case ISD::UNDEF:        SplitRes_UNDEF(N, Lo, Hi); break;
   case ISD::BITCAST:           SplitVecRes_BITCAST(N, Lo, Hi); break;
   case ISD::BUILD_VECTOR:      SplitVecRes_BUILD_VECTOR(N, Lo, Hi); break;
   case ISD::CONCAT_VECTORS:    SplitVecRes_CONCAT_VECTORS(N, Lo, Hi); break;
   case ISD::EXTRACT_SUBVECTOR: SplitVecRes_EXTRACT_SUBVECTOR(N, Lo, Hi); break;
   case ISD::INSERT_SUBVECTOR:  SplitVecRes_INSERT_SUBVECTOR(N, Lo, Hi); break;

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-vpmerge.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+v -target-abi=ilp32d -riscv-v-vector-bits-min=128 \
+; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+v,+m -target-abi=ilp32d -riscv-v-vector-bits-min=128 \
 ; RUN:   -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
-; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+v -target-abi=lp64d -riscv-v-vector-bits-min=128 \
+; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+v,+m -target-abi=lp64d -riscv-v-vector-bits-min=128 \
 ; RUN:   -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
 
 declare <2 x i8> @llvm.vp.merge.v2i8(<2 x i1>, <2 x i8>, <2 x i8>, i32)
 
 define <2 x i8> @vpmerge_vv_v2i8(<2 x i8> %va, <2 x i8> %vb, <2 x i1> %m, i32 zeroext %evl) {
 ; CHECK-LABEL: vpmerge_vv_v2i8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, tu, mu

 ; CHECK-NEXT:    vsetvli zero, a0, e64, m8, tu, mu
 ; CHECK-NEXT:    vfmerge.vfm v8, v8, fa0, v0
 ; CHECK-NEXT:    ret
   %elt.head = insertelement <16 x double> poison, double %a, i32 0
   %va = shufflevector <16 x double> %elt.head, <16 x double> poison, <16 x i32> zeroinitializer
   %v = call <16 x double> @llvm.vp.merge.v16f64(<16 x i1> %m, <16 x double> %va, <16 x double> %vb, i32 %evl)
   ret <16 x double> %v
 }
+
+declare <32 x double> @llvm.vp.merge.v32f64(<32 x i1>, <32 x double>, <32 x double>, i32)
+
+define <32 x double> @vpmerge_vv_v32f64(<32 x double> %va, <32 x double> %vb, <32 x i1> %m, i32 zeroext %evl) {
+; RV32-LABEL: vpmerge_vv_v32f64:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    .cfi_def_cfa_offset 16
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    li a3, 24
+; RV32-NEXT:    mul a1, a1, a3
+; RV32-NEXT:    sub sp, sp, a1
+; RV32-NEXT:    addi a1, a0, 128
+; RV32-NEXT:    vsetivli zero, 16, e64, m8, ta, mu
+; RV32-NEXT:    vle64.v v24, (a1)
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    slli a1, a1, 3
+; RV32-NEXT:    add a1, sp, a1
+; RV32-NEXT:    addi a1, a1, 16
+; RV32-NEXT:    vs8r.v v24, (a1) # Unknown-size Folded Spill
+; RV32-NEXT:    addi a3, a2, -16
+; RV32-NEXT:    vmv1r.v v1, v0
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    slli a1, a1, 4
+; RV32-NEXT:    add a1, sp, a1
+; RV32-NEXT:    addi a1, a1, 16
+; RV32-NEXT:    vs8r.v v16, (a1) # Unknown-size Folded Spill
+; RV32-NEXT:    addi a1, sp, 16
+; RV32-NEXT:    vs8r.v v8, (a1) # Unknown-size Folded Spill
+; RV32-NEXT:    li a1, 0
+; RV32-NEXT:    bltu a2, a3, .LBB72_2
+; RV32-NEXT:  # %bb.1:
+; RV32-NEXT:    mv a1, a3
+; RV32-NEXT:  .LBB72_2:
+; RV32-NEXT:    vle64.v v8, (a0)
+; RV32-NEXT:    vsetivli zero, 2, e8, mf4, ta, mu
+; RV32-NEXT:    vslidedown.vi v0, v1, 2
+; RV32-NEXT:    vsetvli zero, a1, e64, m8, tu, mu
+; RV32-NEXT:    li a0, 16
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    slli a1, a1, 4
+; RV32-NEXT:    add a1, sp, a1
+; RV32-NEXT:    addi a1, a1, 16
+; RV32-NEXT:    vl8re8.v v24, (a1) # Unknown-size Folded Reload
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    slli a1, a1, 3
+; RV32-NEXT:    add a1, sp, a1
+; RV32-NEXT:    addi a1, a1, 16
+; RV32-NEXT:    vl8re8.v v16, (a1) # Unknown-size Folded Reload
+; RV32-NEXT:    vmerge.vvm v16, v16, v24, v0
+; RV32-NEXT:    bltu a2, a0, .LBB72_4
+; RV32-NEXT:  # %bb.3:
+; RV32-NEXT:    li a2, 16
+; RV32-NEXT:  .LBB72_4:
+; RV32-NEXT:    vsetvli zero, a2, e64, m8, tu, mu
+; RV32-NEXT:    vmv1r.v v0, v1
+; RV32-NEXT:    addi a0, sp, 16
+; RV32-NEXT:    vl8re8.v v24, (a0) # Unknown-size Folded Reload
+; RV32-NEXT:    vmerge.vvm v8, v8, v24, v0
+; RV32-NEXT:    csrr a0, vlenb
+; RV32-NEXT:    li a1, 24
+; RV32-NEXT:    mul a0, a0, a1
+; RV32-NEXT:    add sp, sp, a0
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: vpmerge_vv_v32f64:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    .cfi_def_cfa_offset 16
+; RV64-NEXT:    csrr a1, vlenb
+; RV64-NEXT:    slli a1, a1, 4
+; RV64-NEXT:    sub sp, sp, a1
+; RV64-NEXT:    addi a1, a0, 128
+; RV64-NEXT:    vsetivli zero, 16, e64, m8, ta, mu
+; RV64-NEXT:    vle64.v v24, (a1)
+; RV64-NEXT:    addi a3, a2, -16
+; RV64-NEXT:    addi a1, sp, 16
+; RV64-NEXT:    vs8r.v v16, (a1) # Unknown-size Folded Spill
+; RV64-NEXT:    vmv1r.v v1, v0
+; RV64-NEXT:    csrr a1, vlenb
+; RV64-NEXT:    slli a1, a1, 3
+; RV64-NEXT:    add a1, sp, a1
+; RV64-NEXT:    addi a1, a1, 16
+; RV64-NEXT:    vs8r.v v8, (a1) # Unknown-size Folded Spill
+; RV64-NEXT:    li a1, 0
+; RV64-NEXT:    bltu a2, a3, .LBB72_2
+; RV64-NEXT:  # %bb.1:
+; RV64-NEXT:    mv a1, a3
+; RV64-NEXT:  .LBB72_2:
+; RV64-NEXT:    vle64.v v8, (a0)
+; RV64-NEXT:    vsetivli zero, 2, e8, mf4, ta, mu
+; RV64-NEXT:    vslidedown.vi v0, v1, 2
+; RV64-NEXT:    vsetvli zero, a1, e64, m8, tu, mu
+; RV64-NEXT:    li a0, 16
+; RV64-NEXT:    addi a1, sp, 16
+; RV64-NEXT:    vl8re8.v v16, (a1) # Unknown-size Folded Reload
+; RV64-NEXT:    vmerge.vvm v24, v24, v16, v0
+; RV64-NEXT:    bltu a2, a0, .LBB72_4
+; RV64-NEXT:  # %bb.3:
+; RV64-NEXT:    li a2, 16
+; RV64-NEXT:  .LBB72_4:
+; RV64-NEXT:    vsetvli zero, a2, e64, m8, tu, mu
+; RV64-NEXT:    vmv1r.v v0, v1
+; RV64-NEXT:    csrr a0, vlenb
+; RV64-NEXT:    slli a0, a0, 3
+; RV64-NEXT:    add a0, sp, a0
+; RV64-NEXT:    addi a0, a0, 16
+; RV64-NEXT:    vl8re8.v v16, (a0) # Unknown-size Folded Reload
+; RV64-NEXT:    vmerge.vvm v8, v8, v16, v0
+; RV64-NEXT:    vmv8r.v v16, v24
+; RV64-NEXT:    csrr a0, vlenb
+; RV64-NEXT:    slli a0, a0, 4
+; RV64-NEXT:    add sp, sp, a0
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    ret
+  %v = call <32 x double> @llvm.vp.merge.v32f64(<32 x i1> %m, <32 x double> %va, <32 x double> %vb, i32 %evl)
+  ret <32 x double> %v
+}
+
+define <32 x double> @vpmerge_vf_v32f64(double %a, <32 x double> %vb, <32 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vpmerge_vf_v32f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a2, a0, -16
+; CHECK-NEXT:    vmv1r.v v24, v0
+; CHECK-NEXT:    li a1, 0
+; CHECK-NEXT:    bltu a0, a2, .LBB73_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a1, a2
+; CHECK-NEXT:  .LBB73_2:
+; CHECK-NEXT:    vsetivli zero, 2, e8, mf4, ta, mu
+; CHECK-NEXT:    vslidedown.vi v0, v24, 2
+; CHECK-NEXT:    vsetvli zero, a1, e64, m8, tu, mu
+; CHECK-NEXT:    li a1, 16
+; CHECK-NEXT:    vfmerge.vfm v16, v16, fa0, v0
+; CHECK-NEXT:    bltu a0, a1, .LBB73_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    li a0, 16
+; CHECK-NEXT:  .LBB73_4:
+; CHECK-NEXT:    vsetvli zero, a0, e64, m8, tu, mu
+; CHECK-NEXT:    vmv1r.v v0, v24
+; CHECK-NEXT:    vfmerge.vfm v8, v8, fa0, v0
+; CHECK-NEXT:    ret
+  %elt.head = insertelement <32 x double> poison, double %a, i32 0
+  %va = shufflevector <32 x double> %elt.head, <32 x double> poison, <32 x i32> zeroinitializer
+  %v = call <32 x double> @llvm.vp.merge.v32f64(<32 x i1> %m, <32 x double> %va, <32 x double> %vb, i32 %evl)
+  ret <32 x double> %v
+}

llvm/test/CodeGen/RISCV/rvv/vpmerge-sdnode.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+v -target-abi=ilp32d \
+; RUN: llc -mtriple=riscv32 -mattr=+d,+zfh,+v,+m -target-abi=ilp32d \
 ; RUN:     -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV32
-; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+v -target-abi=lp64d \
+; RUN: llc -mtriple=riscv64 -mattr=+d,+zfh,+v,+m -target-abi=lp64d \
 ; RUN:     -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,RV64
 
 declare <vscale x 1 x i8> @llvm.vp.merge.nxv1i8(<vscale x 1 x i1>, <vscale x 1 x i8>, <vscale x 1 x i8>, i32)
 
 define <vscale x 1 x i8> @vpmerge_vv_nxv1i8(<vscale x 1 x i8> %va, <vscale x 1 x i8> %vb, <vscale x 1 x i1> %m, i32 zeroext %evl) {
 ; CHECK-LABEL: vpmerge_vv_nxv1i8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e8, mf8, tu, mu

 ; CHECK-NEXT:    vmerge.vim v8, v8, 2, v0
 ; CHECK-NEXT:    ret
   %elt.head = insertelement <vscale x 64 x i8> poison, i8 2, i32 0
   %va = shufflevector <vscale x 64 x i8> %elt.head, <vscale x 64 x i8> poison, <vscale x 64 x i32> zeroinitializer
   %v = call <vscale x 64 x i8> @llvm.vp.merge.nxv64i8(<vscale x 64 x i1> %m, <vscale x 64 x i8> %va, <vscale x 64 x i8> %vb, i32 %evl)
   ret <vscale x 64 x i8> %v
 }
 
+declare <vscale x 128 x i8> @llvm.vp.merge.nxv128i8(<vscale x 128 x i1>, <vscale x 128 x i8>, <vscale x 128 x i8>, i32)
+
+define <vscale x 128 x i8> @vpmerge_vv_nxv128i8(<vscale x 128 x i8> %va, <vscale x 128 x i8> %vb, <vscale x 128 x i1> %m, i32 zeroext %evl) {
+; RV32-LABEL: vpmerge_vv_nxv128i8:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    .cfi_def_cfa_offset 16
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    li a4, 24
+; RV32-NEXT:    mul a1, a1, a4
+; RV32-NEXT:    sub sp, sp, a1
+; RV32-NEXT:    csrr a1, vlenb
+; RV32-NEXT:    slli a1, a1, 3
+; RV32-NEXT:    add a4, a0, a1
+; RV32-NEXT:    vl8r.v v24, (a4)
+; RV32-NEXT:    csrr a4, vlenb
+; RV32-NEXT:    slli a4, a4, 3
+; RV32-NEXT:    add a4, sp, a4
+; RV32-NEXT:    addi a4, a4, 16
+; RV32-NEXT:    vs8r.v v24, (a4) # Unknown-size Folded Spill
+; RV32-NEXT:    vsetvli a4, zero, e8, m8, ta, mu
+; RV32-NEXT:    vlm.v v2, (a2)
+; RV32-NEXT:    sub a4, a3, a1
+; RV32-NEXT:    vmv1r.v v1, v0
+; RV32-NEXT:    csrr a2, vlenb
+; RV32-NEXT:    slli a2, a2, 4
+; RV32-NEXT:    add a2, sp, a2
+; RV32-NEXT:    addi a2, a2, 16
+; RV32-NEXT:    vs8r.v v16, (a2) # Unknown-size Folded Spill
+; RV32-NEXT:    addi a2, sp, 16
+; RV32-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
+; RV32-NEXT:    li a2, 0
+; RV32-NEXT:    bltu a3, a4, .LBB21_2
+; RV32-NEXT:  # %bb.1:
+; RV32-NEXT:    mv a2, a4
+; RV32-NEXT:  .LBB21_2:
+; RV32-NEXT:    vl8r.v v8, (a0)
+; RV32-NEXT:    vsetvli zero, a2, e8, m8, tu, mu
+; RV32-NEXT:    vmv1r.v v0, v2
+; RV32-NEXT:    csrr a0, vlenb
+; RV32-NEXT:    slli a0, a0, 4
+; RV32-NEXT:    add a0, sp, a0
+; RV32-NEXT:    addi a0, a0, 16
+; RV32-NEXT:    vl8re8.v v24, (a0) # Unknown-size Folded Reload
+; RV32-NEXT:    csrr a0, vlenb
+; RV32-NEXT:    slli a0, a0, 3
+; RV32-NEXT:    add a0, sp, a0
+; RV32-NEXT:    addi a0, a0, 16
+; RV32-NEXT:    vl8re8.v v16, (a0) # Unknown-size Folded Reload
+; RV32-NEXT:    vmerge.vvm v16, v16, v24, v0
+; RV32-NEXT:    bltu a3, a1, .LBB21_4
+; RV32-NEXT:  # %bb.3:
+; RV32-NEXT:    mv a3, a1
+; RV32-NEXT:  .LBB21_4:
+; RV32-NEXT:    vsetvli zero, a3, e8, m8, tu, mu
+; RV32-NEXT:    vmv1r.v v0, v1
+; RV32-NEXT:    addi a0, sp, 16
+; RV32-NEXT:    vl8re8.v v24, (a0) # Unknown-size Folded Reload
+; RV32-NEXT:    vmerge.vvm v8, v8, v24, v0
+; RV32-NEXT:    csrr a0, vlenb
+; RV32-NEXT:    li a1, 24
+; RV32-NEXT:    mul a0, a0, a1
+; RV32-NEXT:    add sp, sp, a0
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: vpmerge_vv_nxv128i8:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    .cfi_def_cfa_offset 16
+; RV64-NEXT:    csrr a1, vlenb
+; RV64-NEXT:    slli a1, a1, 3
+; RV64-NEXT:    sub sp, sp, a1
+; RV64-NEXT:    csrr a1, vlenb
+; RV64-NEXT:    slli a1, a1, 3
+; RV64-NEXT:    add a4, a0, a1
+; RV64-NEXT:    vl8r.v v24, (a4)
+; RV64-NEXT:    vsetvli a4, zero, e8, m8, ta, mu
+; RV64-NEXT:    vlm.v v2, (a2)
+; RV64-NEXT:    sub a4, a3, a1
+; RV64-NEXT:    vmv1r.v v1, v0
+; RV64-NEXT:    addi a2, sp, 16
+; RV64-NEXT:    vs8r.v v8, (a2) # Unknown-size Folded Spill
+; RV64-NEXT:    li a2, 0
+; RV64-NEXT:    bltu a3, a4, .LBB21_2
+; RV64-NEXT:  # %bb.1:
+; RV64-NEXT:    mv a2, a4
+; RV64-NEXT:  .LBB21_2:
+; RV64-NEXT:    vl8r.v v8, (a0)
+; RV64-NEXT:    vsetvli zero, a2, e8, m8, tu, mu
+; RV64-NEXT:    vmv1r.v v0, v2
+; RV64-NEXT:    vmerge.vvm v24, v24, v16, v0
+; RV64-NEXT:    bltu a3, a1, .LBB21_4
+; RV64-NEXT:  # %bb.3:
+; RV64-NEXT:    mv a3, a1
+; RV64-NEXT:  .LBB21_4:
+; RV64-NEXT:    vsetvli zero, a3, e8, m8, tu, mu
+; RV64-NEXT:    vmv1r.v v0, v1
+; RV64-NEXT:    addi a0, sp, 16
+; RV64-NEXT:    vl8re8.v v16, (a0) # Unknown-size Folded Reload
+; RV64-NEXT:    vmerge.vvm v8, v8, v16, v0
+; RV64-NEXT:    vmv8r.v v16, v24
+; RV64-NEXT:    csrr a0, vlenb
+; RV64-NEXT:    slli a0, a0, 3
+; RV64-NEXT:    add sp, sp, a0
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    ret
+  %v = call <vscale x 128 x i8> @llvm.vp.merge.nxv128i8(<vscale x 128 x i1> %m, <vscale x 128 x i8> %va, <vscale x 128 x i8> %vb, i32 %evl)
+  ret <vscale x 128 x i8> %v
+}
+
+define <vscale x 128 x i8> @vpmerge_vx_nxv128i8(i8 %a, <vscale x 128 x i8> %vb, <vscale x 128 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vpmerge_vx_nxv128i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    csrr a3, vlenb
+; CHECK-NEXT:    slli a3, a3, 3
+; CHECK-NEXT:    mv a4, a2
+; CHECK-NEXT:    bltu a2, a3, .LBB22_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a4, a3
+; CHECK-NEXT:  .LBB22_2:
+; CHECK-NEXT:    li a5, 0
+; CHECK-NEXT:    vsetvli a6, zero, e8, m8, ta, mu
+; CHECK-NEXT:    vlm.v v24, (a1)
+; CHECK-NEXT:    vsetvli zero, a4, e8, m8, tu, mu
+; CHECK-NEXT:    sub a1, a2, a3
+; CHECK-NEXT:    vmerge.vxm v8, v8, a0, v0
+; CHECK-NEXT:    bltu a2, a1, .LBB22_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    mv a5, a1
+; CHECK-NEXT:  .LBB22_4:
+; CHECK-NEXT:    vsetvli zero, a5, e8, m8, tu, mu
+; CHECK-NEXT:    vmv1r.v v0, v24
+; CHECK-NEXT:    vmerge.vxm v16, v16, a0, v0
+; CHECK-NEXT:    ret
+  %elt.head = insertelement <vscale x 128 x i8> poison, i8 %a, i32 0
+  %va = shufflevector <vscale x 128 x i8> %elt.head, <vscale x 128 x i8> poison, <vscale x 128 x i32> zeroinitializer
+  %v = call <vscale x 128 x i8> @llvm.vp.merge.nxv128i8(<vscale x 128 x i1> %m, <vscale x 128 x i8> %va, <vscale x 128 x i8> %vb, i32 %evl)
+  ret <vscale x 128 x i8> %v
+}
+
+define <vscale x 128 x i8> @vpmerge_vi_nxv128i8(<vscale x 128 x i8> %vb, <vscale x 128 x i1> %m, i32 zeroext %evl) {
+; CHECK-LABEL: vpmerge_vi_nxv128i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    csrr a2, vlenb
+; CHECK-NEXT:    slli a2, a2, 3
+; CHECK-NEXT:    mv a3, a1
+; CHECK-NEXT:    bltu a1, a2, .LBB23_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    mv a3, a2
+; CHECK-NEXT:  .LBB23_2:
+; CHECK-NEXT:    li a4, 0
+; CHECK-NEXT:    vsetvli a5, zero, e8, m8, ta, mu
+; CHECK-NEXT:    vlm.v v24, (a0)
+; CHECK-NEXT:    vsetvli zero, a3, e8, m8, tu, mu
+; CHECK-NEXT:    sub a0, a1, a2
+; CHECK-NEXT:    vmerge.vim v8, v8, 2, v0
+; CHECK-NEXT:    bltu a1, a0, .LBB23_4
+; CHECK-NEXT:  # %bb.3:
+; CHECK-NEXT:    mv a4, a0
+; CHECK-NEXT:  .LBB23_4:
+; CHECK-NEXT:    vsetvli zero, a4, e8, m8, tu, mu
+; CHECK-NEXT:    vmv1r.v v0, v24
+; CHECK-NEXT:    vmerge.vim v16, v16, 2, v0
+; CHECK-NEXT:    ret
+  %elt.head = insertelement <vscale x 128 x i8> poison, i8 2, i32 0
+  %va = shufflevector <vscale x 128 x i8> %elt.head, <vscale x 128 x i8> poison, <vscale x 128 x i32> zeroinitializer
+  %v = call <vscale x 128 x i8> @llvm.vp.merge.nxv128i8(<vscale x 128 x i1> %m, <vscale x 128 x i8> %va, <vscale x 128 x i8> %vb, i32 %evl)
+  ret <vscale x 128 x i8> %v
+}
+
 declare <vscale x 1 x i16> @llvm.vp.merge.nxv1i16(<vscale x 1 x i1>, <vscale x 1 x i16>, <vscale x 1 x i16>, i32)
 
 define <vscale x 1 x i16> @vpmerge_vv_nxv1i16(<vscale x 1 x i16> %va, <vscale x 1 x i16> %vb, <vscale x 1 x i1> %m, i32 zeroext %evl) {
 ; CHECK-LABEL: vpmerge_vv_nxv1i16:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vsetvli zero, a0, e16, mf4, tu, mu
 ; CHECK-NEXT:    vmerge.vvm v9, v9, v8, v0
 ; CHECK-NEXT:    vmv1r.v v8, v9