[llvm-branch-commits] [llvm] [LoongArch] Perform DAG combine for MUL to generate `[x]vmulw{ev/od}` (PR #161368)

via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Sun Nov 8 19:08:04 PST 2026


https://github.com/zhaoqi5 updated https://github.com/llvm/llvm-project/pull/161368

>From 76630276dcac01a3b37c5278de38d7fbe4023d78 Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Tue, 30 Sep 2025 20:53:06 +0800
Subject: [PATCH 1/7] [LoongArch] Perform DAG combine for MUL to generate
 `[x]vmulw{ev/od}`

---
 .../LoongArch/LoongArchISelLowering.cpp       |  113 ++
 .../LoongArch/LoongArchLASXInstrInfo.td       |   41 +
 .../Target/LoongArch/LoongArchLSXInstrInfo.td |   50 +
 .../lasx/ir-instruction/mulwev_od.ll          | 1040 +----------------
 .../LoongArch/lsx/ir-instruction/mulwev_od.ll |  258 +---
 5 files changed, 247 insertions(+), 1255 deletions(-)

diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 5f266b11149bba..53a2766ca0e1e5 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -542,6 +542,7 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
     setTargetDAGCombine(ISD::UINT_TO_FP);
     setTargetDAGCombine(ISD::ZERO_EXTEND);
     setTargetDAGCombine(ISD::SIGN_EXTEND);
+    setTargetDAGCombine(ISD::MUL);
   }
 
   // Set DAG combine for 'LASX' feature.
@@ -9081,6 +9082,115 @@ static SDValue performVSELECTCombine(SDNode *N, SelectionDAG &DAG,
                      DL, VT, X, Shift);
 }
 
+static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
+                                 TargetLowering::DAGCombinerInfo &DCI) {
+  if (!DCI.isBeforeLegalize())
+    return SDValue();
+
+  SDLoc DL(N);
+  EVT ResTy = N->getValueType(0);
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+
+  if (ResTy != MVT::v8i16 && ResTy != MVT::v4i32 && ResTy != MVT::v2i64 &&
+      ResTy != MVT::v16i16 && ResTy != MVT::v8i32 &&
+      ResTy != MVT::v4i64) // && ResTy != MVT::v2i128)
+    return SDValue();
+
+  // Combine:
+  //   ti,tii,...,tx = extract_vector_elt t0, {0,2,4,.../1,3,5,...}
+  //   tj,tjj,...,ty = extract_vector_elt t1, {0,2,4,.../1,3,5,...}
+  //   tm = BUILD_VECTOR ti,tii,...,tx
+  //   tn = BUILD_VECTOR tj,tjj,...,ty
+  //   ta = {sign/zero}_extend tm
+  //   tb = {sign/zero}_extend tn
+  //   tr = mul ta, tb
+  // to:
+  //   tr = VMULW{EV/OD}[U/US] t0, t1
+  auto getExtType = [](unsigned Op0, unsigned Op1) -> unsigned {
+    if (Op0 == ISD::SIGN_EXTEND && Op1 == ISD::SIGN_EXTEND)
+      return 0;
+    if (Op0 == ISD::ZERO_EXTEND && Op1 == ISD::ZERO_EXTEND)
+      return 1;
+    if (Op0 == ISD::ZERO_EXTEND && Op1 == ISD::SIGN_EXTEND)
+      return 2;
+    if (Op0 == ISD::SIGN_EXTEND && Op1 == ISD::ZERO_EXTEND)
+      return 3;
+    return -1;
+  };
+
+  unsigned ExtType = getExtType(N0.getOpcode(), N1.getOpcode());
+  if (ExtType < 0)
+    return SDValue();
+
+  SDValue BV0 = N0.getOperand(0);
+  SDValue BV1 = N1.getOperand(0);
+  if (BV0.getOpcode() != ISD::BUILD_VECTOR ||
+      BV1.getOpcode() != ISD::BUILD_VECTOR)
+    return SDValue();
+
+  unsigned ResBits = ResTy.getScalarType().getSizeInBits();
+  unsigned BV0Bits = BV0.getValueType().getScalarType().getSizeInBits();
+  unsigned BV1Bits = BV1.getValueType().getScalarType().getSizeInBits();
+  if (BV0Bits != BV1Bits || ResBits != BV0Bits * 2)
+    return SDValue();
+
+  unsigned Index;
+  SDValue OrigN0, OrigN1;
+  for (unsigned i = 0; i < BV0.getNumOperands(); ++i) {
+    SDValue Op0 = BV0.getOperand(i);
+    SDValue Op1 = BV1.getOperand(i);
+    // Each element of BUILD_VECTOR must be EXTRACT_VECTOR_ELT.
+    if (Op0.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
+        Op1.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
+      return SDValue();
+
+    // Check each EXTRACT_VECTOR_ELT's source vector and index.
+    if (Op0.getOperand(1) != Op1.getOperand(1))
+      return SDValue();
+
+    auto *IdxC = dyn_cast<ConstantSDNode>(Op0.getOperand(1));
+    if (!IdxC)
+      return SDValue();
+    unsigned CurIdx = IdxC->getZExtValue();
+
+    if (i == 0) {
+      if (CurIdx != 0 && CurIdx != 1)
+        return SDValue();
+      OrigN0 = Op0.getOperand(0);
+      OrigN1 = Op1.getOperand(0);
+    } else {
+      if (CurIdx != Index + 2)
+        return SDValue();
+      if (Op0.getOperand(0) != OrigN0 || Op1.getOperand(0) != OrigN1)
+        return SDValue();
+    }
+    Index = CurIdx;
+  }
+
+  if (OrigN0.getValueType() != OrigN1.getValueType())
+    return SDValue();
+  if (OrigN0.getValueType().getVectorNumElements() !=
+      ResTy.getVectorNumElements() * 2)
+    return SDValue();
+
+  SDValue Result;
+  EVT OrigTy = OrigN0.getValueType();
+  bool IsEven = (Index % 2 == 0);
+
+  static const unsigned OpcTable[3][2] = {
+      {LoongArchISD::VMULWOD, LoongArchISD::VMULWEV},
+      {LoongArchISD::VMULWODU, LoongArchISD::VMULWEVU},
+      {LoongArchISD::VMULWODUS, LoongArchISD::VMULWEVUS}};
+
+  if (ExtType == 3)
+    Result = DAG.getNode(OpcTable[2][IsEven], DL, OrigTy, OrigN1, OrigN0);
+  else
+    Result = DAG.getNode(OpcTable[ExtType][IsEven], DL, OrigTy, OrigN0, OrigN1);
+
+  return DAG.getBitcast(ResTy, Result);
+}
+
 SDValue LoongArchTargetLowering::PerformDAGCombine(SDNode *N,
                                                    DAGCombinerInfo &DCI) const {
   SelectionDAG &DAG = DCI.DAG;
@@ -9148,6 +9258,9 @@ SDValue LoongArchTargetLowering::PerformDAGCombine(SDNode *N,
     if (SDValue Result =
             combineFP_ROUND(SDValue(N, 0), SDLoc(N), DAG, Subtarget))
       return Result;
+    break;
+  case ISD::MUL:
+    return performMULCombine(N, DAG, DCI);
   }
   return SDValue();
 }
diff --git a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
index 6b0a426ed1d1f4..07a5a9a8dcaa88 100644
--- a/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchLASXInstrInfo.td
@@ -2009,6 +2009,39 @@ multiclass PairInsertExtractPatV4<ValueType vecty, ValueType elemty> {
   }
 }
 
+multiclass XVmulwPat<SDPatternOperator OpNode, string Inst> {
+  def : Pat<(OpNode (v32i8 LASX256:$xj), (v32i8 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_H_B") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v16i16 LASX256:$xj), (v16i16 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_W_H") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v8i32 LASX256:$xj), (v8i32 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_D_W") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v4i64 LASX256:$xj), (v4i64 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_Q_D") LASX256:$xj, LASX256:$xk)>;
+}
+
+multiclass XVmulwuPat<SDPatternOperator OpNode, string Inst> {
+  def : Pat<(OpNode (v32i8 LASX256:$xj), (v32i8 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_H_BU") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v16i16 LASX256:$xj), (v16i16 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_W_HU") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v8i32 LASX256:$xj), (v8i32 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_D_WU") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v4i64 LASX256:$xj), (v4i64 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_Q_DU") LASX256:$xj, LASX256:$xk)>;
+}
+
+multiclass XVmulwusPat<SDPatternOperator OpNode, string Inst> {
+  def : Pat<(OpNode (v32i8 LASX256:$xj), (v32i8 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_H_BU_B") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v16i16 LASX256:$xj), (v16i16 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_W_HU_H") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v8i32 LASX256:$xj), (v8i32 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_D_WU_W") LASX256:$xj, LASX256:$xk)>;
+  def : Pat<(OpNode (v4i64 LASX256:$xj), (v4i64 LASX256:$xk)),
+            (!cast<LAInst>(Inst#"_Q_DU_D") LASX256:$xj, LASX256:$xk)>;
+}
+
 let Predicates = [HasExtLASX] in {
 
 // XVADD_{B/H/W/D}
@@ -2048,6 +2081,14 @@ defm : PatXrXr<mul, "XVMUL">;
 defm : PatXrXr<mulhs, "XVMUH">;
 defm : PatXrXrU<mulhu, "XVMUH">;
 
+// XVMULW{EV/OD}_{H_B/W_H/D_W/Q_D}[U], XVMULW{EV/OD}_{H_BU_B/W_HU_H/D_WU_W/Q_DU_D}
+defm : XVmulwPat<loongarch_vmulwev, "XVMULWEV">;
+defm : XVmulwPat<loongarch_vmulwod, "XVMULWOD">;
+defm : XVmulwuPat<loongarch_vmulwevu, "XVMULWEV">;
+defm : XVmulwuPat<loongarch_vmulwodu, "XVMULWOD">;
+defm : XVmulwusPat<loongarch_vmulwevus, "XVMULWEV">;
+defm : XVmulwusPat<loongarch_vmulwodus, "XVMULWOD">;
+
 // XVMADD_{B/H/W/D}
 defm : PatXrXrXr<muladd, "XVMADD">;
 // XVMSUB_{B/H/W/D}
diff --git a/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
index eb0a2a7f5f5530..1e2fc69b00efcd 100644
--- a/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
@@ -139,6 +139,15 @@ def loongarch_vanynonzero
 def loongarch_vallzero
     : SDNode<"LoongArchISD::VALLZERO", SDT_LoongArchVMSKCOND>;
 
+// Perform element-wise vector multiplication at even/odd indices,
+// and keep each result in its corresponding widened slot
+def loongarch_vmulwev: SDNode<"LoongArchISD::VMULWEV", SDT_LoongArchV2R>;
+def loongarch_vmulwod: SDNode<"LoongArchISD::VMULWOD", SDT_LoongArchV2R>;
+def loongarch_vmulwevu: SDNode<"LoongArchISD::VMULWEVU", SDT_LoongArchV2R>;
+def loongarch_vmulwodu: SDNode<"LoongArchISD::VMULWODU", SDT_LoongArchV2R>;
+def loongarch_vmulwevus: SDNode<"LoongArchISD::VMULWEVUS", SDT_LoongArchV2R>;
+def loongarch_vmulwodus: SDNode<"LoongArchISD::VMULWODUS", SDT_LoongArchV2R>;
+
 def immZExt1 : ImmLeaf<GRLenVT, [{return isUInt<1>(Imm);}]>;
 def immZExt2 : ImmLeaf<GRLenVT, [{return isUInt<2>(Imm);}]>;
 def immZExt3 : ImmLeaf<GRLenVT, [{return isUInt<3>(Imm);}]>;
@@ -2244,6 +2253,39 @@ multiclass InsertExtractPatV2<ValueType vecty, ValueType elemty> {
   }
 }
 
+multiclass VmulwPat<SDPatternOperator OpNode, string Inst> {
+  def : Pat<(OpNode (v16i8 LSX128:$vj), (v16i8 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_H_B") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v8i16 LSX128:$vj), (v8i16 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_W_H") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v4i32 LSX128:$vj), (v4i32 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_D_W") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v2i64 LSX128:$vj), (v2i64 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_Q_D") LSX128:$vj, LSX128:$vk)>;
+}
+
+multiclass VmulwuPat<SDPatternOperator OpNode, string Inst> {
+  def : Pat<(OpNode (v16i8 LSX128:$vj), (v16i8 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_H_BU") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v8i16 LSX128:$vj), (v8i16 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_W_HU") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v4i32 LSX128:$vj), (v4i32 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_D_WU") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v2i64 LSX128:$vj), (v2i64 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_Q_DU") LSX128:$vj, LSX128:$vk)>;
+}
+
+multiclass VmulwusPat<SDPatternOperator OpNode, string Inst> {
+  def : Pat<(OpNode (v16i8 LSX128:$vj), (v16i8 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_H_BU_B") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v8i16 LSX128:$vj), (v8i16 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_W_HU_H") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v4i32 LSX128:$vj), (v4i32 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_D_WU_W") LSX128:$vj, LSX128:$vk)>;
+  def : Pat<(OpNode (v2i64 LSX128:$vj), (v2i64 LSX128:$vk)),
+            (!cast<LAInst>(Inst#"_Q_DU_D") LSX128:$vj, LSX128:$vk)>;
+}
+
 let Predicates = [HasExtLSX] in {
 
 // VADD_{B/H/W/D}
@@ -2283,6 +2325,14 @@ defm : PatVrVr<mul, "VMUL">;
 defm : PatVrVr<mulhs, "VMUH">;
 defm : PatVrVrU<mulhu, "VMUH">;
 
+// VMULW{EV/OD}_{H_B/W_H/D_W/Q_D}[U], VMULW{EV/OD}_{H_BU_B/W_HU_H/D_WU_W/Q_DU_D}
+defm : VmulwPat<loongarch_vmulwev, "VMULWEV">;
+defm : VmulwPat<loongarch_vmulwod, "VMULWOD">;
+defm : VmulwuPat<loongarch_vmulwevu, "VMULWEV">;
+defm : VmulwuPat<loongarch_vmulwodu, "VMULWOD">;
+defm : VmulwusPat<loongarch_vmulwevus, "VMULWEV">;
+defm : VmulwusPat<loongarch_vmulwodus, "VMULWOD">;
+
 // VMADD_{B/H/W/D}
 defm : PatVrVrVr<muladd, "VMADD">;
 // VMSUB_{B/H/W/D}
diff --git a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll
index 2b8ca39bff3859..605325f4dc4f4b 100644
--- a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll
@@ -5,77 +5,9 @@
 define void @vmulwev_h_b(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_h_b:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.h.b $xr0, $xr0
-; CHECK-NEXT:    vext2xv.h.b $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwev.h.b $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -95,43 +27,7 @@ define void @vmulwev_w_h(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.w.h $xr1, $xr2
-; CHECK-NEXT:    vext2xv.w.h $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.w.h $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -151,25 +47,7 @@ define void @vmulwev_d_w(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.d.w $xr1, $xr2
-; CHECK-NEXT:    vext2xv.d.w $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.d.w $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -333,77 +211,9 @@ entry:
 define void @vmulwod_h_b(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwod_h_b:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.h.b $xr0, $xr0
-; CHECK-NEXT:    vext2xv.h.b $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwod.h.b $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -423,43 +233,7 @@ define void @vmulwod_w_h(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.w.h $xr1, $xr2
-; CHECK-NEXT:    vext2xv.w.h $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.w.h $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -479,25 +253,7 @@ define void @vmulwod_d_w(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.d.w $xr1, $xr2
-; CHECK-NEXT:    vext2xv.d.w $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.d.w $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -661,77 +417,9 @@ entry:
 define void @vmulwev_h_bu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_h_bu:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.hu.bu $xr0, $xr0
-; CHECK-NEXT:    vext2xv.hu.bu $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwev.h.bu $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -751,43 +439,7 @@ define void @vmulwev_w_hu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.wu.hu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.wu.hu $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.w.hu $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -807,25 +459,7 @@ define void @vmulwev_d_wu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.du.wu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.du.wu $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.d.wu $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -935,77 +569,9 @@ entry:
 define void @vmulwod_h_bu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwod_h_bu:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.hu.bu $xr0, $xr0
-; CHECK-NEXT:    vext2xv.hu.bu $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwod.h.bu $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1025,43 +591,7 @@ define void @vmulwod_w_hu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.wu.hu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.wu.hu $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.w.hu $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1081,25 +611,7 @@ define void @vmulwod_d_wu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.du.wu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.du.wu $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.d.wu $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1209,77 +721,9 @@ entry:
 define void @vmulwev_h_bu_b(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_h_bu_b:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.hu.bu $xr0, $xr0
-; CHECK-NEXT:    vext2xv.h.b $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwev.h.bu.b $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1299,43 +743,7 @@ define void @vmulwev_w_hu_h(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.wu.hu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.w.h $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.w.hu.h $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1355,25 +763,7 @@ define void @vmulwev_d_wu_w(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.du.wu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.d.w $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.d.wu.w $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1515,77 +905,9 @@ entry:
 define void @vmulwod_h_bu_b(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwod_h_bu_b:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.hu.bu $xr0, $xr0
-; CHECK-NEXT:    vext2xv.h.b $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwod.h.bu.b $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1605,43 +927,7 @@ define void @vmulwod_w_hu_h(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.wu.hu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.w.h $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.w.hu.h $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1661,25 +947,7 @@ define void @vmulwod_d_wu_w(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.du.wu $xr1, $xr2
-; CHECK-NEXT:    vext2xv.d.w $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.d.wu.w $xr0, $xr0, $xr1
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1821,77 +1089,9 @@ entry:
 define void @vmulwev_h_bu_b_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_h_bu_b_1:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 8
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 10
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 12
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 14
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.h.b $xr0, $xr0
-; CHECK-NEXT:    vext2xv.hu.bu $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwev.h.bu.b $xr0, $xr1, $xr0
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1911,43 +1111,7 @@ define void @vmulwev_w_hu_h_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 0
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 2
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 4
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 6
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.w.h $xr1, $xr2
-; CHECK-NEXT:    vext2xv.wu.hu $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.w.hu.h $xr0, $xr1, $xr0
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -1967,25 +1131,7 @@ define void @vmulwev_d_wu_w_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 0
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 2
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 4
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 6
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.d.w $xr1, $xr2
-; CHECK-NEXT:    vext2xv.du.wu $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwev.d.wu.w $xr0, $xr1, $xr0
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -2127,77 +1273,9 @@ entry:
 define void @vmulwod_h_bu_b_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwod_h_bu_b_1:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    xvld $xr2, $a1, 0
+; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr2, $xr2, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr2, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr0, $a1, 15
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 3
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 7
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 8
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 9
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 10
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 11
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 9
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 12
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 11
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 13
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 13
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 14
-; CHECK-NEXT:    vpickve2gr.b $a1, $vr1, 15
-; CHECK-NEXT:    vinsgr2vr.b $vr2, $a1, 15
-; CHECK-NEXT:    vext2xv.h.b $xr0, $xr0
-; CHECK-NEXT:    vext2xv.hu.bu $xr1, $xr2
-; CHECK-NEXT:    xvmul.h $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvmulwod.h.bu.b $xr0, $xr1, $xr0
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -2217,43 +1295,7 @@ define void @vmulwod_w_hu_h_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr0, $xr0, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr0, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr2, $a1, 7
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 0
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 1
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 2
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 3
-; CHECK-NEXT:    xvpermi.d $xr1, $xr1, 14
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 1
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 4
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 3
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 5
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 5
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 6
-; CHECK-NEXT:    vpickve2gr.h $a1, $vr1, 7
-; CHECK-NEXT:    vinsgr2vr.h $vr0, $a1, 7
-; CHECK-NEXT:    vext2xv.w.h $xr1, $xr2
-; CHECK-NEXT:    vext2xv.wu.hu $xr0, $xr0
-; CHECK-NEXT:    xvmul.w $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.w.hu.h $xr0, $xr1, $xr0
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -2273,25 +1315,7 @@ define void @vmulwod_d_wu_w_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xvld $xr0, $a1, 0
 ; CHECK-NEXT:    xvld $xr1, $a2, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr0, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr2, $a1, 3
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 1
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 0
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 3
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 1
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 5
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 2
-; CHECK-NEXT:    xvpickve2gr.w $a1, $xr1, 7
-; CHECK-NEXT:    vinsgr2vr.w $vr0, $a1, 3
-; CHECK-NEXT:    vext2xv.d.w $xr1, $xr2
-; CHECK-NEXT:    vext2xv.du.wu $xr0, $xr0
-; CHECK-NEXT:    xvmul.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvmulwod.d.wu.w $xr0, $xr1, $xr0
 ; CHECK-NEXT:    xvst $xr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
index dfad02660718fe..293f6f341e7fe1 100644
--- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
@@ -3,17 +3,6 @@
 ; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA64
 
 define void @vmulwev_h_b(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_h_b:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vpickev.b $vr0, $vr0, $vr0
-; CHECK-NEXT:    vpickev.b $vr1, $vr1, $vr1
-; CHECK-NEXT:    vsllwil.h.b $vr0, $vr0, 0
-; CHECK-NEXT:    vsllwil.h.b $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -27,17 +16,6 @@ entry:
 }
 
 define void @vmulwev_w_h(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_w_h:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vpickev.h $vr0, $vr0, $vr0
-; CHECK-NEXT:    vpickev.h $vr1, $vr1, $vr1
-; CHECK-NEXT:    vsllwil.w.h $vr0, $vr0, 0
-; CHECK-NEXT:    vsllwil.w.h $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.w $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -51,17 +29,6 @@ entry:
 }
 
 define void @vmulwev_d_w(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_d_w:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vshuf4i.w $vr0, $vr0, 8
-; CHECK-NEXT:    vshuf4i.w $vr1, $vr1, 8
-; CHECK-NEXT:    vsllwil.d.w $vr0, $vr0, 0
-; CHECK-NEXT:    vsllwil.d.w $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.d $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -151,17 +118,6 @@ entry:
 }
 
 define void @vmulwod_h_b(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_h_b:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vpickod.b $vr0, $vr0, $vr0
-; CHECK-NEXT:    vpickod.b $vr1, $vr1, $vr1
-; CHECK-NEXT:    vsllwil.h.b $vr0, $vr0, 0
-; CHECK-NEXT:    vsllwil.h.b $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -175,17 +131,6 @@ entry:
 }
 
 define void @vmulwod_w_h(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_w_h:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vpickod.h $vr0, $vr0, $vr0
-; CHECK-NEXT:    vpickod.h $vr1, $vr1, $vr1
-; CHECK-NEXT:    vsllwil.w.h $vr0, $vr0, 0
-; CHECK-NEXT:    vsllwil.w.h $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.w $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -199,17 +144,6 @@ entry:
 }
 
 define void @vmulwod_d_w(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_d_w:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vshuf4i.w $vr0, $vr0, 13
-; CHECK-NEXT:    vshuf4i.w $vr1, $vr1, 13
-; CHECK-NEXT:    vsllwil.d.w $vr0, $vr0, 0
-; CHECK-NEXT:    vsllwil.d.w $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.d $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -302,13 +236,8 @@ define void @vmulwev_h_bu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_h_bu:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI8_0)
-; CHECK-NEXT:    vld $vr1, $a1, %pc_lo12(.LCPI8_0)
-; CHECK-NEXT:    vld $vr2, $a2, 0
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.b $vr0, $vr3, $vr0, $vr1
-; CHECK-NEXT:    vshuf.b $vr1, $vr3, $vr2, $vr1
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.h.bu $vr0, $vr0, $vr1
 ; CHECK-NEXT:    vst $vr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -326,15 +255,9 @@ entry:
 define void @vmulwev_w_hu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_w_hu:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    pcalau12i $a3, %pc_hi20(.LCPI9_0)
-; CHECK-NEXT:    vld $vr0, $a3, %pc_lo12(.LCPI9_0)
-; CHECK-NEXT:    vld $vr1, $a1, 0
-; CHECK-NEXT:    vld $vr2, $a2, 0
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vori.b $vr4, $vr0, 0
-; CHECK-NEXT:    vshuf.h $vr4, $vr3, $vr1
-; CHECK-NEXT:    vshuf.h $vr0, $vr3, $vr2
-; CHECK-NEXT:    vmul.w $vr0, $vr4, $vr0
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.w.hu $vr0, $vr0, $vr1
 ; CHECK-NEXT:    vst $vr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -352,15 +275,9 @@ entry:
 define void @vmulwev_d_wu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_d_wu:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    pcalau12i $a3, %pc_hi20(.LCPI10_0)
-; CHECK-NEXT:    vld $vr0, $a3, %pc_lo12(.LCPI10_0)
-; CHECK-NEXT:    vld $vr1, $a1, 0
-; CHECK-NEXT:    vld $vr2, $a2, 0
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vori.b $vr4, $vr0, 0
-; CHECK-NEXT:    vshuf.w $vr4, $vr3, $vr1
-; CHECK-NEXT:    vshuf.w $vr0, $vr3, $vr2
-; CHECK-NEXT:    vmul.d $vr0, $vr4, $vr0
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.d.wu $vr0, $vr0, $vr1
 ; CHECK-NEXT:    vst $vr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -436,10 +353,7 @@ define void @vmulwod_h_bu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vld $vr0, $a1, 0
 ; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.b $vr0, $vr2, $vr0
-; CHECK-NEXT:    vpackod.b $vr1, $vr2, $vr1
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
+; CHECK-NEXT:    vmulwod.h.bu $vr0, $vr0, $vr1
 ; CHECK-NEXT:    vst $vr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -459,10 +373,7 @@ define void @vmulwod_w_hu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vld $vr0, $a1, 0
 ; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.h $vr0, $vr2, $vr0
-; CHECK-NEXT:    vpackod.h $vr1, $vr2, $vr1
-; CHECK-NEXT:    vmul.w $vr0, $vr0, $vr1
+; CHECK-NEXT:    vmulwod.w.hu $vr0, $vr0, $vr1
 ; CHECK-NEXT:    vst $vr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -482,10 +393,7 @@ define void @vmulwod_d_wu(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vld $vr0, $a1, 0
 ; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.w $vr0, $vr2, $vr0
-; CHECK-NEXT:    vpackod.w $vr1, $vr2, $vr1
-; CHECK-NEXT:    vmul.d $vr0, $vr0, $vr1
+; CHECK-NEXT:    vmulwod.d.wu $vr0, $vr0, $vr1
 ; CHECK-NEXT:    vst $vr0, $a0, 0
 ; CHECK-NEXT:    ret
 entry:
@@ -557,19 +465,6 @@ entry:
 }
 
 define void @vmulwev_h_bu_b(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_h_bu_b:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI16_0)
-; CHECK-NEXT:    vld $vr2, $a1, %pc_lo12(.LCPI16_0)
-; CHECK-NEXT:    vpickev.b $vr1, $vr1, $vr1
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.b $vr0, $vr3, $vr0, $vr2
-; CHECK-NEXT:    vsllwil.h.b $vr1, $vr1, 0
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -583,19 +478,6 @@ entry:
 }
 
 define void @vmulwev_w_hu_h(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_w_hu_h:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI17_0)
-; CHECK-NEXT:    vld $vr2, $a1, %pc_lo12(.LCPI17_0)
-; CHECK-NEXT:    vpickev.h $vr1, $vr1, $vr1
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.h $vr2, $vr3, $vr0
-; CHECK-NEXT:    vsllwil.w.h $vr0, $vr1, 0
-; CHECK-NEXT:    vmul.w $vr0, $vr2, $vr0
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -609,19 +491,6 @@ entry:
 }
 
 define void @vmulwev_d_wu_w(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_d_wu_w:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI18_0)
-; CHECK-NEXT:    vld $vr2, $a1, %pc_lo12(.LCPI18_0)
-; CHECK-NEXT:    vshuf4i.w $vr1, $vr1, 8
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.w $vr2, $vr3, $vr0
-; CHECK-NEXT:    vsllwil.d.w $vr0, $vr1, 0
-; CHECK-NEXT:    vmul.d $vr0, $vr2, $vr0
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -704,17 +573,6 @@ entry:
 }
 
 define void @vmulwod_h_bu_b(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_h_bu_b:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a2, 0
-; CHECK-NEXT:    vld $vr1, $a1, 0
-; CHECK-NEXT:    vpickod.b $vr0, $vr0, $vr0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.b $vr1, $vr2, $vr1
-; CHECK-NEXT:    vsllwil.h.b $vr0, $vr0, 0
-; CHECK-NEXT:    vmul.h $vr0, $vr1, $vr0
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -728,17 +586,6 @@ entry:
 }
 
 define void @vmulwod_w_hu_h(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_w_hu_h:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a2, 0
-; CHECK-NEXT:    vld $vr1, $a1, 0
-; CHECK-NEXT:    vpickod.h $vr0, $vr0, $vr0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.h $vr1, $vr2, $vr1
-; CHECK-NEXT:    vsllwil.w.h $vr0, $vr0, 0
-; CHECK-NEXT:    vmul.w $vr0, $vr1, $vr0
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -752,17 +599,6 @@ entry:
 }
 
 define void @vmulwod_d_wu_w(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_d_wu_w:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a2, 0
-; CHECK-NEXT:    vld $vr1, $a1, 0
-; CHECK-NEXT:    vshuf4i.w $vr0, $vr0, 13
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.w $vr1, $vr2, $vr1
-; CHECK-NEXT:    vsllwil.d.w $vr0, $vr0, 0
-; CHECK-NEXT:    vmul.d $vr0, $vr1, $vr0
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -845,19 +681,6 @@ entry:
 }
 
 define void @vmulwev_h_bu_b_1(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_h_bu_b_1:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI24_0)
-; CHECK-NEXT:    vld $vr2, $a1, %pc_lo12(.LCPI24_0)
-; CHECK-NEXT:    vpickev.b $vr0, $vr0, $vr0
-; CHECK-NEXT:    vsllwil.h.b $vr0, $vr0, 0
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.b $vr1, $vr3, $vr1, $vr2
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -871,19 +694,6 @@ entry:
 }
 
 define void @vmulwev_w_hu_h_1(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_w_hu_h_1:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI25_0)
-; CHECK-NEXT:    vld $vr2, $a1, %pc_lo12(.LCPI25_0)
-; CHECK-NEXT:    vpickev.h $vr0, $vr0, $vr0
-; CHECK-NEXT:    vsllwil.w.h $vr0, $vr0, 0
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.h $vr2, $vr3, $vr1
-; CHECK-NEXT:    vmul.w $vr0, $vr0, $vr2
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -897,19 +707,6 @@ entry:
 }
 
 define void @vmulwev_d_wu_w_1(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwev_d_wu_w_1:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    pcalau12i $a1, %pc_hi20(.LCPI26_0)
-; CHECK-NEXT:    vld $vr2, $a1, %pc_lo12(.LCPI26_0)
-; CHECK-NEXT:    vshuf4i.w $vr0, $vr0, 8
-; CHECK-NEXT:    vsllwil.d.w $vr0, $vr0, 0
-; CHECK-NEXT:    vrepli.b $vr3, 0
-; CHECK-NEXT:    vshuf.w $vr2, $vr3, $vr1
-; CHECK-NEXT:    vmul.d $vr0, $vr0, $vr2
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -992,17 +789,6 @@ entry:
 }
 
 define void @vmulwod_h_bu_b_1(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_h_bu_b_1:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vpickod.b $vr0, $vr0, $vr0
-; CHECK-NEXT:    vsllwil.h.b $vr0, $vr0, 0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.b $vr1, $vr2, $vr1
-; CHECK-NEXT:    vmul.h $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -1016,17 +802,6 @@ entry:
 }
 
 define void @vmulwod_w_hu_h_1(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_w_hu_h_1:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vpickod.h $vr0, $vr0, $vr0
-; CHECK-NEXT:    vsllwil.w.h $vr0, $vr0, 0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.h $vr1, $vr2, $vr1
-; CHECK-NEXT:    vmul.w $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -1040,17 +815,6 @@ entry:
 }
 
 define void @vmulwod_d_wu_w_1(ptr %res, ptr %a, ptr %b) nounwind {
-; CHECK-LABEL: vmulwod_d_wu_w_1:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vld $vr0, $a1, 0
-; CHECK-NEXT:    vld $vr1, $a2, 0
-; CHECK-NEXT:    vshuf4i.w $vr0, $vr0, 13
-; CHECK-NEXT:    vsllwil.d.w $vr0, $vr0, 0
-; CHECK-NEXT:    vrepli.b $vr2, 0
-; CHECK-NEXT:    vpackod.w $vr1, $vr2, $vr1
-; CHECK-NEXT:    vmul.d $vr0, $vr0, $vr1
-; CHECK-NEXT:    vst $vr0, $a0, 0
-; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b

>From b8ff6cadf7e6050024aab12480e1973846e402bc Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Thu, 9 Oct 2025 15:59:19 +0800
Subject: [PATCH 2/7] deal with lsx i128

---
 .../LoongArch/LoongArchISelLowering.cpp       |  56 ++-
 .../LoongArch/lsx/ir-instruction/mulwev_od.ll | 452 ++++--------------
 2 files changed, 130 insertions(+), 378 deletions(-)

diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 53a2766ca0e1e5..3283bf75e3bf62 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -9093,15 +9093,15 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
   SDValue N1 = N->getOperand(1);
 
   if (ResTy != MVT::v8i16 && ResTy != MVT::v4i32 && ResTy != MVT::v2i64 &&
-      ResTy != MVT::v16i16 && ResTy != MVT::v8i32 &&
-      ResTy != MVT::v4i64) // && ResTy != MVT::v2i128)
+      ResTy != MVT::v16i16 && ResTy != MVT::v8i32 && ResTy != MVT::v4i64 &&
+      ResTy != MVT::i128)
     return SDValue();
 
   // Combine:
   //   ti,tii,...,tx = extract_vector_elt t0, {0,2,4,.../1,3,5,...}
   //   tj,tjj,...,ty = extract_vector_elt t1, {0,2,4,.../1,3,5,...}
-  //   tm = BUILD_VECTOR ti,tii,...,tx
-  //   tn = BUILD_VECTOR tj,tjj,...,ty
+  //   tm = BUILD_VECTOR ti,tii,...,tx (Only when ResTy != MVT::i128)
+  //   tn = BUILD_VECTOR tj,tjj,...,ty (Only when ResTy != MVT::i128)
   //   ta = {sign/zero}_extend tm
   //   tb = {sign/zero}_extend tn
   //   tr = mul ta, tb
@@ -9123,24 +9123,36 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
   if (ExtType < 0)
     return SDValue();
 
-  SDValue BV0 = N0.getOperand(0);
-  SDValue BV1 = N1.getOperand(0);
-  if (BV0.getOpcode() != ISD::BUILD_VECTOR ||
-      BV1.getOpcode() != ISD::BUILD_VECTOR)
+  SDValue Src0 = N0.getOperand(0);
+  SDValue Src1 = N1.getOperand(0);
+  bool IsScalar = (ResTy == MVT::i128);
+  if (IsScalar && (Src0.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
+                   Src1.getOpcode() != ISD::EXTRACT_VECTOR_ELT))
+    return SDValue();
+  if (!IsScalar && (Src0.getOpcode() != ISD::BUILD_VECTOR ||
+                    Src1.getOpcode() != ISD::BUILD_VECTOR))
     return SDValue();
 
-  unsigned ResBits = ResTy.getScalarType().getSizeInBits();
-  unsigned BV0Bits = BV0.getValueType().getScalarType().getSizeInBits();
-  unsigned BV1Bits = BV1.getValueType().getScalarType().getSizeInBits();
-  if (BV0Bits != BV1Bits || ResBits != BV0Bits * 2)
+  unsigned ResBits = ResTy.getScalarSizeInBits();
+  unsigned Src0Bits = Src0.getValueType().getScalarSizeInBits();
+  unsigned Src1Bits = Src1.getValueType().getScalarSizeInBits();
+  if (Src0Bits != Src1Bits || ResBits != Src0Bits * 2)
     return SDValue();
 
+  // Collect all EXTRACT_VECTOR_ELT.
+  SmallVector<std::pair<SDValue, SDValue>> Elems;
+  if (IsScalar) {
+    Elems.emplace_back(Src0, Src1);
+  } else {
+    for (unsigned i = 0; i < Src0.getNumOperands(); ++i)
+      Elems.emplace_back(Src0.getOperand(i), Src1.getOperand(i));
+  }
+
   unsigned Index;
   SDValue OrigN0, OrigN1;
-  for (unsigned i = 0; i < BV0.getNumOperands(); ++i) {
-    SDValue Op0 = BV0.getOperand(i);
-    SDValue Op1 = BV1.getOperand(i);
-    // Each element of BUILD_VECTOR must be EXTRACT_VECTOR_ELT.
+  bool First = true;
+  for (auto &[Op0, Op1] : Elems) {
+    // Each element must be EXTRACT_VECTOR_ELT.
     if (Op0.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
         Op1.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
       return SDValue();
@@ -9152,17 +9164,17 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
     auto *IdxC = dyn_cast<ConstantSDNode>(Op0.getOperand(1));
     if (!IdxC)
       return SDValue();
-    unsigned CurIdx = IdxC->getZExtValue();
 
-    if (i == 0) {
+    unsigned CurIdx = IdxC->getZExtValue();
+    if (First) {
       if (CurIdx != 0 && CurIdx != 1)
         return SDValue();
       OrigN0 = Op0.getOperand(0);
       OrigN1 = Op1.getOperand(0);
+      First = false;
     } else {
-      if (CurIdx != Index + 2)
-        return SDValue();
-      if (Op0.getOperand(0) != OrigN0 || Op1.getOperand(0) != OrigN1)
+      if (CurIdx != Index + 2 || Op0.getOperand(0) != OrigN0 ||
+          Op1.getOperand(0) != OrigN1)
         return SDValue();
     }
     Index = CurIdx;
@@ -9171,7 +9183,7 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
   if (OrigN0.getValueType() != OrigN1.getValueType())
     return SDValue();
   if (OrigN0.getValueType().getVectorNumElements() !=
-      ResTy.getVectorNumElements() * 2)
+      (IsScalar ? 1 : ResTy.getVectorNumElements()) * 2)
     return SDValue();
 
   SDValue Result;
diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
index 293f6f341e7fe1..e765898ff58482 100644
--- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
@@ -46,64 +46,23 @@ define void @vmulwev_q_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
+; LA32-NEXT:    vmulwev.q.d $vr0, $vr0, $vr1
 ; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
 ; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 0
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 1
-; LA32-NEXT:    srai.w $a5, $a2, 31
-; LA32-NEXT:    srai.w $a6, $a4, 31
-; LA32-NEXT:    mulh.wu $a7, $a1, $a3
-; LA32-NEXT:    mul.w $t0, $a2, $a3
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    sltu $t0, $a7, $t0
-; LA32-NEXT:    mulh.wu $t1, $a2, $a3
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    mul.w $t1, $a1, $a4
-; LA32-NEXT:    add.w $a7, $t1, $a7
-; LA32-NEXT:    sltu $t1, $a7, $t1
-; LA32-NEXT:    mulh.wu $t2, $a1, $a4
-; LA32-NEXT:    add.w $t1, $t2, $t1
-; LA32-NEXT:    add.w $t1, $t0, $t1
-; LA32-NEXT:    mul.w $t2, $a2, $a4
-; LA32-NEXT:    add.w $t3, $t2, $t1
-; LA32-NEXT:    mul.w $t4, $a3, $a5
-; LA32-NEXT:    mul.w $t5, $a6, $a1
-; LA32-NEXT:    add.w $t6, $t5, $t4
-; LA32-NEXT:    add.w $t7, $t3, $t6
-; LA32-NEXT:    sltu $t8, $t7, $t3
-; LA32-NEXT:    sltu $t2, $t3, $t2
-; LA32-NEXT:    sltu $t0, $t1, $t0
-; LA32-NEXT:    mulh.wu $t1, $a2, $a4
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    add.w $t0, $t0, $t2
-; LA32-NEXT:    mulh.wu $t1, $a3, $a5
-; LA32-NEXT:    add.w $t1, $t1, $t4
-; LA32-NEXT:    mul.w $a4, $a4, $a5
-; LA32-NEXT:    add.w $a4, $t1, $a4
-; LA32-NEXT:    mul.w $a2, $a6, $a2
-; LA32-NEXT:    mulh.wu $a5, $a6, $a1
-; LA32-NEXT:    add.w $a2, $a5, $a2
-; LA32-NEXT:    add.w $a2, $a2, $t5
-; LA32-NEXT:    add.w $a2, $a2, $a4
-; LA32-NEXT:    sltu $a4, $t6, $t5
-; LA32-NEXT:    add.w $a2, $a2, $a4
-; LA32-NEXT:    add.w $a2, $t0, $a2
-; LA32-NEXT:    add.w $a2, $a2, $t8
-; LA32-NEXT:    mul.w $a1, $a1, $a3
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
 ; LA32-NEXT:    st.w $a1, $a0, 0
-; LA32-NEXT:    st.w $a7, $a0, 4
-; LA32-NEXT:    st.w $t7, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 12
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_d:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a1, $a1, 0
-; LA64-NEXT:    ld.d $a2, $a2, 0
-; LA64-NEXT:    mul.d $a3, $a1, $a2
-; LA64-NEXT:    mulh.d $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 8
-; LA64-NEXT:    st.d $a3, $a0, 0
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwev.q.d $vr0, $vr0, $vr1
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -161,64 +120,23 @@ define void @vmulwod_q_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 3
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 3
-; LA32-NEXT:    srai.w $a5, $a2, 31
-; LA32-NEXT:    srai.w $a6, $a4, 31
-; LA32-NEXT:    mulh.wu $a7, $a1, $a3
-; LA32-NEXT:    mul.w $t0, $a2, $a3
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    sltu $t0, $a7, $t0
-; LA32-NEXT:    mulh.wu $t1, $a2, $a3
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    mul.w $t1, $a1, $a4
-; LA32-NEXT:    add.w $a7, $t1, $a7
-; LA32-NEXT:    sltu $t1, $a7, $t1
-; LA32-NEXT:    mulh.wu $t2, $a1, $a4
-; LA32-NEXT:    add.w $t1, $t2, $t1
-; LA32-NEXT:    add.w $t1, $t0, $t1
-; LA32-NEXT:    mul.w $t2, $a2, $a4
-; LA32-NEXT:    add.w $t3, $t2, $t1
-; LA32-NEXT:    mul.w $t4, $a3, $a5
-; LA32-NEXT:    mul.w $t5, $a6, $a1
-; LA32-NEXT:    add.w $t6, $t5, $t4
-; LA32-NEXT:    add.w $t7, $t3, $t6
-; LA32-NEXT:    sltu $t8, $t7, $t3
-; LA32-NEXT:    sltu $t2, $t3, $t2
-; LA32-NEXT:    sltu $t0, $t1, $t0
-; LA32-NEXT:    mulh.wu $t1, $a2, $a4
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    add.w $t0, $t0, $t2
-; LA32-NEXT:    mulh.wu $t1, $a3, $a5
-; LA32-NEXT:    add.w $t1, $t1, $t4
-; LA32-NEXT:    mul.w $a4, $a4, $a5
-; LA32-NEXT:    add.w $a4, $t1, $a4
-; LA32-NEXT:    mul.w $a2, $a6, $a2
-; LA32-NEXT:    mulh.wu $a5, $a6, $a1
-; LA32-NEXT:    add.w $a2, $a5, $a2
-; LA32-NEXT:    add.w $a2, $a2, $t5
-; LA32-NEXT:    add.w $a2, $a2, $a4
-; LA32-NEXT:    sltu $a4, $t6, $t5
-; LA32-NEXT:    add.w $a2, $a2, $a4
-; LA32-NEXT:    add.w $a2, $t0, $a2
-; LA32-NEXT:    add.w $a2, $a2, $t8
-; LA32-NEXT:    mul.w $a1, $a1, $a3
+; LA32-NEXT:    vmulwod.q.d $vr0, $vr0, $vr1
+; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
+; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
 ; LA32-NEXT:    st.w $a1, $a0, 0
-; LA32-NEXT:    st.w $a7, $a0, 4
-; LA32-NEXT:    st.w $t7, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 12
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_d:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a1, $a1, 8
-; LA64-NEXT:    ld.d $a2, $a2, 8
-; LA64-NEXT:    mul.d $a3, $a1, $a2
-; LA64-NEXT:    mulh.d $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 8
-; LA64-NEXT:    st.d $a3, $a0, 0
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwod.q.d $vr0, $vr0, $vr1
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -297,44 +215,23 @@ define void @vmulwev_q_du(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 1
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 0
-; LA32-NEXT:    mulh.wu $a5, $a2, $a4
-; LA32-NEXT:    mul.w $a6, $a1, $a4
-; LA32-NEXT:    add.w $a5, $a6, $a5
-; LA32-NEXT:    sltu $a6, $a5, $a6
-; LA32-NEXT:    mulh.wu $a7, $a1, $a4
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    mul.w $a7, $a2, $a3
-; LA32-NEXT:    add.w $a5, $a7, $a5
-; LA32-NEXT:    sltu $a7, $a5, $a7
-; LA32-NEXT:    mulh.wu $t0, $a2, $a3
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    add.w $a7, $a6, $a7
-; LA32-NEXT:    mul.w $t0, $a1, $a3
-; LA32-NEXT:    add.w $t1, $t0, $a7
-; LA32-NEXT:    sltu $t0, $t1, $t0
-; LA32-NEXT:    sltu $a6, $a7, $a6
-; LA32-NEXT:    mulh.wu $a1, $a1, $a3
-; LA32-NEXT:    add.w $a1, $a1, $a6
-; LA32-NEXT:    add.w $a1, $a1, $t0
-; LA32-NEXT:    mul.w $a2, $a2, $a4
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $a5, $a0, 4
-; LA32-NEXT:    st.w $t1, $a0, 8
-; LA32-NEXT:    st.w $a1, $a0, 12
+; LA32-NEXT:    vmulwev.q.du $vr0, $vr0, $vr1
+; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
+; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
+; LA32-NEXT:    st.w $a1, $a0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_du:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a1, $a1, 0
-; LA64-NEXT:    ld.d $a2, $a2, 0
-; LA64-NEXT:    mul.d $a3, $a1, $a2
-; LA64-NEXT:    mulh.du $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 8
-; LA64-NEXT:    st.d $a3, $a0, 0
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwev.q.du $vr0, $vr0, $vr1
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -413,44 +310,23 @@ define void @vmulwod_q_du(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 3
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 3
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 2
-; LA32-NEXT:    mulh.wu $a5, $a2, $a4
-; LA32-NEXT:    mul.w $a6, $a1, $a4
-; LA32-NEXT:    add.w $a5, $a6, $a5
-; LA32-NEXT:    sltu $a6, $a5, $a6
-; LA32-NEXT:    mulh.wu $a7, $a1, $a4
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    mul.w $a7, $a2, $a3
-; LA32-NEXT:    add.w $a5, $a7, $a5
-; LA32-NEXT:    sltu $a7, $a5, $a7
-; LA32-NEXT:    mulh.wu $t0, $a2, $a3
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    add.w $a7, $a6, $a7
-; LA32-NEXT:    mul.w $t0, $a1, $a3
-; LA32-NEXT:    add.w $t1, $t0, $a7
-; LA32-NEXT:    sltu $t0, $t1, $t0
-; LA32-NEXT:    sltu $a6, $a7, $a6
-; LA32-NEXT:    mulh.wu $a1, $a1, $a3
-; LA32-NEXT:    add.w $a1, $a1, $a6
-; LA32-NEXT:    add.w $a1, $a1, $t0
-; LA32-NEXT:    mul.w $a2, $a2, $a4
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $a5, $a0, 4
-; LA32-NEXT:    st.w $t1, $a0, 8
-; LA32-NEXT:    st.w $a1, $a0, 12
+; LA32-NEXT:    vmulwod.q.du $vr0, $vr0, $vr1
+; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
+; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
+; LA32-NEXT:    st.w $a1, $a0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_du:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a1, $a1, 8
-; LA64-NEXT:    ld.d $a2, $a2, 8
-; LA64-NEXT:    mul.d $a3, $a1, $a2
-; LA64-NEXT:    mulh.du $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 8
-; LA64-NEXT:    st.d $a3, $a0, 0
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwod.q.du $vr0, $vr0, $vr1
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -508,57 +384,23 @@ define void @vmulwev_q_du_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 0
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 1
-; LA32-NEXT:    srai.w $a5, $a4, 31
-; LA32-NEXT:    mulh.wu $a6, $a2, $a3
-; LA32-NEXT:    mul.w $a7, $a1, $a3
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    sltu $a7, $a6, $a7
-; LA32-NEXT:    mulh.wu $t0, $a1, $a3
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    mul.w $t0, $a2, $a4
-; LA32-NEXT:    add.w $a6, $t0, $a6
-; LA32-NEXT:    sltu $t0, $a6, $t0
-; LA32-NEXT:    mulh.wu $t1, $a2, $a4
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    add.w $t0, $a7, $t0
-; LA32-NEXT:    mul.w $t1, $a1, $a4
-; LA32-NEXT:    add.w $t2, $t1, $t0
-; LA32-NEXT:    mul.w $t3, $a5, $a2
-; LA32-NEXT:    add.w $t4, $t2, $t3
-; LA32-NEXT:    sltu $t5, $t4, $t2
-; LA32-NEXT:    sltu $t1, $t2, $t1
-; LA32-NEXT:    sltu $a7, $t0, $a7
-; LA32-NEXT:    mulh.wu $a4, $a1, $a4
-; LA32-NEXT:    add.w $a4, $a4, $a7
-; LA32-NEXT:    add.w $a4, $a4, $t1
-; LA32-NEXT:    mul.w $a1, $a5, $a1
-; LA32-NEXT:    mulh.wu $a5, $a5, $a2
-; LA32-NEXT:    add.w $a1, $a5, $a1
-; LA32-NEXT:    add.w $a1, $a1, $t3
-; LA32-NEXT:    add.w $a1, $a4, $a1
-; LA32-NEXT:    add.w $a1, $a1, $t5
-; LA32-NEXT:    mul.w $a2, $a2, $a3
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $a6, $a0, 4
-; LA32-NEXT:    st.w $t4, $a0, 8
-; LA32-NEXT:    st.w $a1, $a0, 12
+; LA32-NEXT:    vmulwev.q.du.d $vr0, $vr0, $vr1
+; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
+; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
+; LA32-NEXT:    st.w $a1, $a0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_du_d:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a2, $a2, 0
-; LA64-NEXT:    ld.d $a1, $a1, 0
-; LA64-NEXT:    srai.d $a3, $a2, 63
-; LA64-NEXT:    mulh.du $a4, $a1, $a2
-; LA64-NEXT:    mul.d $a3, $a1, $a3
-; LA64-NEXT:    add.d $a3, $a4, $a3
-; LA64-NEXT:    mul.d $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 0
-; LA64-NEXT:    st.d $a3, $a0, 8
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwev.q.du.d $vr0, $vr0, $vr1
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -616,57 +458,23 @@ define void @vmulwod_q_du_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 3
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 3
-; LA32-NEXT:    srai.w $a5, $a4, 31
-; LA32-NEXT:    mulh.wu $a6, $a2, $a3
-; LA32-NEXT:    mul.w $a7, $a1, $a3
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    sltu $a7, $a6, $a7
-; LA32-NEXT:    mulh.wu $t0, $a1, $a3
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    mul.w $t0, $a2, $a4
-; LA32-NEXT:    add.w $a6, $t0, $a6
-; LA32-NEXT:    sltu $t0, $a6, $t0
-; LA32-NEXT:    mulh.wu $t1, $a2, $a4
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    add.w $t0, $a7, $t0
-; LA32-NEXT:    mul.w $t1, $a1, $a4
-; LA32-NEXT:    add.w $t2, $t1, $t0
-; LA32-NEXT:    mul.w $t3, $a5, $a2
-; LA32-NEXT:    add.w $t4, $t2, $t3
-; LA32-NEXT:    sltu $t5, $t4, $t2
-; LA32-NEXT:    sltu $t1, $t2, $t1
-; LA32-NEXT:    sltu $a7, $t0, $a7
-; LA32-NEXT:    mulh.wu $a4, $a1, $a4
-; LA32-NEXT:    add.w $a4, $a4, $a7
-; LA32-NEXT:    add.w $a4, $a4, $t1
-; LA32-NEXT:    mul.w $a1, $a5, $a1
-; LA32-NEXT:    mulh.wu $a5, $a5, $a2
-; LA32-NEXT:    add.w $a1, $a5, $a1
-; LA32-NEXT:    add.w $a1, $a1, $t3
-; LA32-NEXT:    add.w $a1, $a4, $a1
-; LA32-NEXT:    add.w $a1, $a1, $t5
-; LA32-NEXT:    mul.w $a2, $a2, $a3
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $a6, $a0, 4
-; LA32-NEXT:    st.w $t4, $a0, 8
-; LA32-NEXT:    st.w $a1, $a0, 12
+; LA32-NEXT:    vmulwod.q.du.d $vr0, $vr0, $vr1
+; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
+; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
+; LA32-NEXT:    st.w $a1, $a0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_du_d:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a2, $a2, 8
-; LA64-NEXT:    ld.d $a1, $a1, 8
-; LA64-NEXT:    srai.d $a3, $a2, 63
-; LA64-NEXT:    mulh.du $a4, $a1, $a2
-; LA64-NEXT:    mul.d $a3, $a1, $a3
-; LA64-NEXT:    add.d $a3, $a4, $a3
-; LA64-NEXT:    mul.d $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 0
-; LA64-NEXT:    st.d $a3, $a0, 8
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwod.q.du.d $vr0, $vr0, $vr1
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -724,57 +532,23 @@ define void @vmulwev_q_du_d_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
+; LA32-NEXT:    vmulwev.q.du.d $vr0, $vr1, $vr0
 ; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
 ; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 1
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 0
-; LA32-NEXT:    srai.w $a5, $a2, 31
-; LA32-NEXT:    mulh.wu $a6, $a1, $a4
-; LA32-NEXT:    mul.w $a7, $a2, $a4
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    sltu $a7, $a6, $a7
-; LA32-NEXT:    mulh.wu $t0, $a2, $a4
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    mul.w $t0, $a1, $a3
-; LA32-NEXT:    add.w $a6, $t0, $a6
-; LA32-NEXT:    sltu $t0, $a6, $t0
-; LA32-NEXT:    mulh.wu $t1, $a1, $a3
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    add.w $t0, $a7, $t0
-; LA32-NEXT:    mul.w $t1, $a2, $a3
-; LA32-NEXT:    add.w $t2, $t1, $t0
-; LA32-NEXT:    mul.w $t3, $a4, $a5
-; LA32-NEXT:    add.w $t4, $t2, $t3
-; LA32-NEXT:    sltu $t5, $t4, $t2
-; LA32-NEXT:    sltu $t1, $t2, $t1
-; LA32-NEXT:    sltu $a7, $t0, $a7
-; LA32-NEXT:    mulh.wu $a2, $a2, $a3
-; LA32-NEXT:    add.w $a2, $a2, $a7
-; LA32-NEXT:    add.w $a2, $a2, $t1
-; LA32-NEXT:    mulh.wu $a7, $a4, $a5
-; LA32-NEXT:    add.w $a7, $a7, $t3
-; LA32-NEXT:    mul.w $a3, $a3, $a5
-; LA32-NEXT:    add.w $a3, $a7, $a3
-; LA32-NEXT:    add.w $a2, $a2, $a3
-; LA32-NEXT:    add.w $a2, $a2, $t5
-; LA32-NEXT:    mul.w $a1, $a1, $a4
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
 ; LA32-NEXT:    st.w $a1, $a0, 0
-; LA32-NEXT:    st.w $a6, $a0, 4
-; LA32-NEXT:    st.w $t4, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 12
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_du_d_1:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a1, $a1, 0
-; LA64-NEXT:    ld.d $a2, $a2, 0
-; LA64-NEXT:    srai.d $a3, $a1, 63
-; LA64-NEXT:    mulh.du $a4, $a1, $a2
-; LA64-NEXT:    mul.d $a3, $a3, $a2
-; LA64-NEXT:    add.d $a3, $a4, $a3
-; LA64-NEXT:    mul.d $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 0
-; LA64-NEXT:    st.d $a3, $a0, 8
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwev.q.du.d $vr0, $vr1, $vr0
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -832,57 +606,23 @@ define void @vmulwod_q_du_d_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 3
-; LA32-NEXT:    vpickve2gr.w $a3, $vr1, 3
-; LA32-NEXT:    vpickve2gr.w $a4, $vr1, 2
-; LA32-NEXT:    srai.w $a5, $a2, 31
-; LA32-NEXT:    mulh.wu $a6, $a1, $a4
-; LA32-NEXT:    mul.w $a7, $a2, $a4
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    sltu $a7, $a6, $a7
-; LA32-NEXT:    mulh.wu $t0, $a2, $a4
-; LA32-NEXT:    add.w $a7, $t0, $a7
-; LA32-NEXT:    mul.w $t0, $a1, $a3
-; LA32-NEXT:    add.w $a6, $t0, $a6
-; LA32-NEXT:    sltu $t0, $a6, $t0
-; LA32-NEXT:    mulh.wu $t1, $a1, $a3
-; LA32-NEXT:    add.w $t0, $t1, $t0
-; LA32-NEXT:    add.w $t0, $a7, $t0
-; LA32-NEXT:    mul.w $t1, $a2, $a3
-; LA32-NEXT:    add.w $t2, $t1, $t0
-; LA32-NEXT:    mul.w $t3, $a4, $a5
-; LA32-NEXT:    add.w $t4, $t2, $t3
-; LA32-NEXT:    sltu $t5, $t4, $t2
-; LA32-NEXT:    sltu $t1, $t2, $t1
-; LA32-NEXT:    sltu $a7, $t0, $a7
-; LA32-NEXT:    mulh.wu $a2, $a2, $a3
-; LA32-NEXT:    add.w $a2, $a2, $a7
-; LA32-NEXT:    add.w $a2, $a2, $t1
-; LA32-NEXT:    mulh.wu $a7, $a4, $a5
-; LA32-NEXT:    add.w $a7, $a7, $t3
-; LA32-NEXT:    mul.w $a3, $a3, $a5
-; LA32-NEXT:    add.w $a3, $a7, $a3
-; LA32-NEXT:    add.w $a2, $a2, $a3
-; LA32-NEXT:    add.w $a2, $a2, $t5
-; LA32-NEXT:    mul.w $a1, $a1, $a4
+; LA32-NEXT:    vmulwod.q.du.d $vr0, $vr1, $vr0
+; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
+; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
+; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
+; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
+; LA32-NEXT:    st.w $a4, $a0, 12
+; LA32-NEXT:    st.w $a3, $a0, 8
+; LA32-NEXT:    st.w $a2, $a0, 4
 ; LA32-NEXT:    st.w $a1, $a0, 0
-; LA32-NEXT:    st.w $a6, $a0, 4
-; LA32-NEXT:    st.w $t4, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 12
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_du_d_1:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    ld.d $a1, $a1, 8
-; LA64-NEXT:    ld.d $a2, $a2, 8
-; LA64-NEXT:    srai.d $a3, $a1, 63
-; LA64-NEXT:    mulh.du $a4, $a1, $a2
-; LA64-NEXT:    mul.d $a3, $a3, $a2
-; LA64-NEXT:    add.d $a3, $a4, $a3
-; LA64-NEXT:    mul.d $a1, $a1, $a2
-; LA64-NEXT:    st.d $a1, $a0, 0
-; LA64-NEXT:    st.d $a3, $a0, 8
+; LA64-NEXT:    vld $vr0, $a1, 0
+; LA64-NEXT:    vld $vr1, $a2, 0
+; LA64-NEXT:    vmulwod.q.du.d $vr0, $vr1, $vr0
+; LA64-NEXT:    vst $vr0, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a

>From 0586e4c6397f704dd85a3d49a6b16cf9f3d0070e Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Thu, 9 Oct 2025 17:21:23 +0800
Subject: [PATCH 3/7] including v2i128

---
 .../LoongArch/LoongArchISelLowering.cpp       |  17 +-
 .../lasx/ir-instruction/mulwev_od.ll          | 928 ++----------------
 2 files changed, 71 insertions(+), 874 deletions(-)

diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 3283bf75e3bf62..8ab812263fed49 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -9083,7 +9083,8 @@ static SDValue performVSELECTCombine(SDNode *N, SelectionDAG &DAG,
 }
 
 static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
-                                 TargetLowering::DAGCombinerInfo &DCI) {
+                                 TargetLowering::DAGCombinerInfo &DCI,
+                                 const LoongArchSubtarget &Subtarget) {
   if (!DCI.isBeforeLegalize())
     return SDValue();
 
@@ -9092,9 +9093,17 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);
 
+  // Note: v2i128 is an unsupported MVT vector type (see
+  // MachineValueType.h::getVectorVT()), use NumElements and SizeInBits to
+  // identify it.
+  bool HasLSXOnly = Subtarget.hasExtLSX() && !Subtarget.hasExtLASX();
+  bool Isv2i128 = ResTy.isVector() && ResTy.getVectorNumElements() == 2 &&
+                  ResTy.getScalarSizeInBits() == 128;
   if (ResTy != MVT::v8i16 && ResTy != MVT::v4i32 && ResTy != MVT::v2i64 &&
-      ResTy != MVT::v16i16 && ResTy != MVT::v8i32 && ResTy != MVT::v4i64 &&
-      ResTy != MVT::i128)
+      ResTy != MVT::i128 && ResTy != MVT::v16i16 && ResTy != MVT::v8i32 &&
+      ResTy != MVT::v4i64 && !Isv2i128)
+    return SDValue();
+  if (HasLSXOnly && (ResTy.is256BitVector() || Isv2i128))
     return SDValue();
 
   // Combine:
@@ -9272,7 +9281,7 @@ SDValue LoongArchTargetLowering::PerformDAGCombine(SDNode *N,
       return Result;
     break;
   case ISD::MUL:
-    return performMULCombine(N, DAG, DCI);
+    return performMULCombine(N, DAG, DCI, Subtarget);
   }
   return SDValue();
 }
diff --git a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll
index 605325f4dc4f4b..ed3a31d12ee833 100644
--- a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/mulwev_od.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
-; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s --check-prefixes=CHECK,LA32
-; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s --check-prefixes=CHECK,LA64
+; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s
+; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s
 
 define void @vmulwev_h_b(ptr %res, ptr %a, ptr %b) nounwind {
 ; CHECK-LABEL: vmulwev_h_b:
@@ -63,139 +63,13 @@ entry:
 }
 
 define void @vmulwev_q_d(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwev_q_d:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $sp, $sp, -32
-; LA32-NEXT:    st.w $fp, $sp, 28 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s0, $sp, 24 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s1, $sp, 20 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s2, $sp, 16 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s3, $sp, 12 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s4, $sp, 8 # 4-byte Folded Spill
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 0
-; LA32-NEXT:    xvpickve2gr.w $t3, $xr0, 1
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr0, 5
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr1, 4
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr1, 0
-; LA32-NEXT:    xvpickve2gr.w $t4, $xr1, 1
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr1, 5
-; LA32-NEXT:    srai.w $t1, $a5, 31
-; LA32-NEXT:    srai.w $t5, $t3, 31
-; LA32-NEXT:    srai.w $t0, $a7, 31
-; LA32-NEXT:    srai.w $t6, $t4, 31
-; LA32-NEXT:    mulh.wu $a6, $a2, $a4
-; LA32-NEXT:    mul.w $t2, $t3, $a4
-; LA32-NEXT:    add.w $a6, $t2, $a6
-; LA32-NEXT:    sltu $t2, $a6, $t2
-; LA32-NEXT:    mulh.wu $t7, $t3, $a4
-; LA32-NEXT:    add.w $t7, $t7, $t2
-; LA32-NEXT:    mul.w $t2, $a2, $t4
-; LA32-NEXT:    add.w $a6, $t2, $a6
-; LA32-NEXT:    sltu $t2, $a6, $t2
-; LA32-NEXT:    mulh.wu $t8, $a2, $t4
-; LA32-NEXT:    add.w $t2, $t8, $t2
-; LA32-NEXT:    add.w $t8, $t7, $t2
-; LA32-NEXT:    mul.w $fp, $t3, $t4
-; LA32-NEXT:    add.w $s0, $fp, $t8
-; LA32-NEXT:    mul.w $s1, $a4, $t5
-; LA32-NEXT:    mul.w $s2, $t6, $a2
-; LA32-NEXT:    add.w $s3, $s2, $s1
-; LA32-NEXT:    add.w $t2, $s0, $s3
-; LA32-NEXT:    sltu $s4, $t2, $s0
-; LA32-NEXT:    sltu $fp, $s0, $fp
-; LA32-NEXT:    sltu $t7, $t8, $t7
-; LA32-NEXT:    mulh.wu $t8, $t3, $t4
-; LA32-NEXT:    add.w $t7, $t8, $t7
-; LA32-NEXT:    add.w $t7, $t7, $fp
-; LA32-NEXT:    mulh.wu $t8, $a4, $t5
-; LA32-NEXT:    add.w $t8, $t8, $s1
-; LA32-NEXT:    mul.w $t4, $t4, $t5
-; LA32-NEXT:    add.w $t4, $t8, $t4
-; LA32-NEXT:    mul.w $t3, $t6, $t3
-; LA32-NEXT:    mulh.wu $t5, $t6, $a2
-; LA32-NEXT:    add.w $t3, $t5, $t3
-; LA32-NEXT:    add.w $t3, $t3, $s2
-; LA32-NEXT:    add.w $t3, $t3, $t4
-; LA32-NEXT:    sltu $t4, $s3, $s2
-; LA32-NEXT:    add.w $t3, $t3, $t4
-; LA32-NEXT:    add.w $t3, $t7, $t3
-; LA32-NEXT:    add.w $t3, $t3, $s4
-; LA32-NEXT:    mulh.wu $t4, $a1, $a3
-; LA32-NEXT:    mul.w $t5, $a5, $a3
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    sltu $t5, $t4, $t5
-; LA32-NEXT:    mulh.wu $t6, $a5, $a3
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    mul.w $t6, $a1, $a7
-; LA32-NEXT:    add.w $t4, $t6, $t4
-; LA32-NEXT:    sltu $t6, $t4, $t6
-; LA32-NEXT:    mulh.wu $t7, $a1, $a7
-; LA32-NEXT:    add.w $t6, $t7, $t6
-; LA32-NEXT:    add.w $t6, $t5, $t6
-; LA32-NEXT:    mul.w $t7, $a5, $a7
-; LA32-NEXT:    add.w $t8, $t7, $t6
-; LA32-NEXT:    mul.w $fp, $a3, $t1
-; LA32-NEXT:    mul.w $s0, $t0, $a1
-; LA32-NEXT:    add.w $s1, $s0, $fp
-; LA32-NEXT:    add.w $s2, $t8, $s1
-; LA32-NEXT:    sltu $s3, $s2, $t8
-; LA32-NEXT:    sltu $t7, $t8, $t7
-; LA32-NEXT:    sltu $t5, $t6, $t5
-; LA32-NEXT:    mulh.wu $t6, $a5, $a7
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    add.w $t5, $t5, $t7
-; LA32-NEXT:    mulh.wu $t6, $a3, $t1
-; LA32-NEXT:    add.w $t6, $t6, $fp
-; LA32-NEXT:    mul.w $a7, $a7, $t1
-; LA32-NEXT:    add.w $a7, $t6, $a7
-; LA32-NEXT:    mul.w $a5, $t0, $a5
-; LA32-NEXT:    mulh.wu $t0, $t0, $a1
-; LA32-NEXT:    add.w $a5, $t0, $a5
-; LA32-NEXT:    add.w $a5, $a5, $s0
-; LA32-NEXT:    add.w $a5, $a5, $a7
-; LA32-NEXT:    sltu $a7, $s1, $s0
-; LA32-NEXT:    add.w $a5, $a5, $a7
-; LA32-NEXT:    add.w $a5, $t5, $a5
-; LA32-NEXT:    add.w $a5, $a5, $s3
-; LA32-NEXT:    mul.w $a2, $a2, $a4
-; LA32-NEXT:    mul.w $a1, $a1, $a3
-; LA32-NEXT:    st.w $a1, $a0, 16
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $t4, $a0, 20
-; LA32-NEXT:    st.w $a6, $a0, 4
-; LA32-NEXT:    st.w $s2, $a0, 24
-; LA32-NEXT:    st.w $t2, $a0, 8
-; LA32-NEXT:    st.w $a5, $a0, 28
-; LA32-NEXT:    st.w $t3, $a0, 12
-; LA32-NEXT:    ld.w $s4, $sp, 8 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s3, $sp, 12 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s2, $sp, 16 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s1, $sp, 20 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s0, $sp, 24 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $fp, $sp, 28 # 4-byte Folded Reload
-; LA32-NEXT:    addi.w $sp, $sp, 32
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwev_q_d:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 2
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 0
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 2
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 0
-; LA64-NEXT:    mul.d $a5, $a2, $a4
-; LA64-NEXT:    mulh.d $a2, $a2, $a4
-; LA64-NEXT:    mul.d $a4, $a1, $a3
-; LA64-NEXT:    mulh.d $a1, $a1, $a3
-; LA64-NEXT:    st.d $a1, $a0, 24
-; LA64-NEXT:    st.d $a4, $a0, 16
-; LA64-NEXT:    st.d $a2, $a0, 8
-; LA64-NEXT:    st.d $a5, $a0, 0
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwev_q_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwev.q.d $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -269,139 +143,13 @@ entry:
 }
 
 define void @vmulwod_q_d(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwod_q_d:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $sp, $sp, -32
-; LA32-NEXT:    st.w $fp, $sp, 28 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s0, $sp, 24 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s1, $sp, 20 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s2, $sp, 16 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s3, $sp, 12 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s4, $sp, 8 # 4-byte Folded Spill
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 2
-; LA32-NEXT:    xvpickve2gr.w $t3, $xr0, 3
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr0, 7
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr1, 6
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr1, 2
-; LA32-NEXT:    xvpickve2gr.w $t4, $xr1, 3
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr1, 7
-; LA32-NEXT:    srai.w $t1, $a5, 31
-; LA32-NEXT:    srai.w $t5, $t3, 31
-; LA32-NEXT:    srai.w $t0, $a7, 31
-; LA32-NEXT:    srai.w $t6, $t4, 31
-; LA32-NEXT:    mulh.wu $a6, $a2, $a4
-; LA32-NEXT:    mul.w $t2, $t3, $a4
-; LA32-NEXT:    add.w $a6, $t2, $a6
-; LA32-NEXT:    sltu $t2, $a6, $t2
-; LA32-NEXT:    mulh.wu $t7, $t3, $a4
-; LA32-NEXT:    add.w $t7, $t7, $t2
-; LA32-NEXT:    mul.w $t2, $a2, $t4
-; LA32-NEXT:    add.w $a6, $t2, $a6
-; LA32-NEXT:    sltu $t2, $a6, $t2
-; LA32-NEXT:    mulh.wu $t8, $a2, $t4
-; LA32-NEXT:    add.w $t2, $t8, $t2
-; LA32-NEXT:    add.w $t8, $t7, $t2
-; LA32-NEXT:    mul.w $fp, $t3, $t4
-; LA32-NEXT:    add.w $s0, $fp, $t8
-; LA32-NEXT:    mul.w $s1, $a4, $t5
-; LA32-NEXT:    mul.w $s2, $t6, $a2
-; LA32-NEXT:    add.w $s3, $s2, $s1
-; LA32-NEXT:    add.w $t2, $s0, $s3
-; LA32-NEXT:    sltu $s4, $t2, $s0
-; LA32-NEXT:    sltu $fp, $s0, $fp
-; LA32-NEXT:    sltu $t7, $t8, $t7
-; LA32-NEXT:    mulh.wu $t8, $t3, $t4
-; LA32-NEXT:    add.w $t7, $t8, $t7
-; LA32-NEXT:    add.w $t7, $t7, $fp
-; LA32-NEXT:    mulh.wu $t8, $a4, $t5
-; LA32-NEXT:    add.w $t8, $t8, $s1
-; LA32-NEXT:    mul.w $t4, $t4, $t5
-; LA32-NEXT:    add.w $t4, $t8, $t4
-; LA32-NEXT:    mul.w $t3, $t6, $t3
-; LA32-NEXT:    mulh.wu $t5, $t6, $a2
-; LA32-NEXT:    add.w $t3, $t5, $t3
-; LA32-NEXT:    add.w $t3, $t3, $s2
-; LA32-NEXT:    add.w $t3, $t3, $t4
-; LA32-NEXT:    sltu $t4, $s3, $s2
-; LA32-NEXT:    add.w $t3, $t3, $t4
-; LA32-NEXT:    add.w $t3, $t7, $t3
-; LA32-NEXT:    add.w $t3, $t3, $s4
-; LA32-NEXT:    mulh.wu $t4, $a1, $a3
-; LA32-NEXT:    mul.w $t5, $a5, $a3
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    sltu $t5, $t4, $t5
-; LA32-NEXT:    mulh.wu $t6, $a5, $a3
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    mul.w $t6, $a1, $a7
-; LA32-NEXT:    add.w $t4, $t6, $t4
-; LA32-NEXT:    sltu $t6, $t4, $t6
-; LA32-NEXT:    mulh.wu $t7, $a1, $a7
-; LA32-NEXT:    add.w $t6, $t7, $t6
-; LA32-NEXT:    add.w $t6, $t5, $t6
-; LA32-NEXT:    mul.w $t7, $a5, $a7
-; LA32-NEXT:    add.w $t8, $t7, $t6
-; LA32-NEXT:    mul.w $fp, $a3, $t1
-; LA32-NEXT:    mul.w $s0, $t0, $a1
-; LA32-NEXT:    add.w $s1, $s0, $fp
-; LA32-NEXT:    add.w $s2, $t8, $s1
-; LA32-NEXT:    sltu $s3, $s2, $t8
-; LA32-NEXT:    sltu $t7, $t8, $t7
-; LA32-NEXT:    sltu $t5, $t6, $t5
-; LA32-NEXT:    mulh.wu $t6, $a5, $a7
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    add.w $t5, $t5, $t7
-; LA32-NEXT:    mulh.wu $t6, $a3, $t1
-; LA32-NEXT:    add.w $t6, $t6, $fp
-; LA32-NEXT:    mul.w $a7, $a7, $t1
-; LA32-NEXT:    add.w $a7, $t6, $a7
-; LA32-NEXT:    mul.w $a5, $t0, $a5
-; LA32-NEXT:    mulh.wu $t0, $t0, $a1
-; LA32-NEXT:    add.w $a5, $t0, $a5
-; LA32-NEXT:    add.w $a5, $a5, $s0
-; LA32-NEXT:    add.w $a5, $a5, $a7
-; LA32-NEXT:    sltu $a7, $s1, $s0
-; LA32-NEXT:    add.w $a5, $a5, $a7
-; LA32-NEXT:    add.w $a5, $t5, $a5
-; LA32-NEXT:    add.w $a5, $a5, $s3
-; LA32-NEXT:    mul.w $a2, $a2, $a4
-; LA32-NEXT:    mul.w $a1, $a1, $a3
-; LA32-NEXT:    st.w $a1, $a0, 16
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $t4, $a0, 20
-; LA32-NEXT:    st.w $a6, $a0, 4
-; LA32-NEXT:    st.w $s2, $a0, 24
-; LA32-NEXT:    st.w $t2, $a0, 8
-; LA32-NEXT:    st.w $a5, $a0, 28
-; LA32-NEXT:    st.w $t3, $a0, 12
-; LA32-NEXT:    ld.w $s4, $sp, 8 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s3, $sp, 12 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s2, $sp, 16 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s1, $sp, 20 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $s0, $sp, 24 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $fp, $sp, 28 # 4-byte Folded Reload
-; LA32-NEXT:    addi.w $sp, $sp, 32
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwod_q_d:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 3
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 1
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 3
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 1
-; LA64-NEXT:    mul.d $a5, $a2, $a4
-; LA64-NEXT:    mulh.d $a2, $a2, $a4
-; LA64-NEXT:    mul.d $a4, $a1, $a3
-; LA64-NEXT:    mulh.d $a1, $a1, $a3
-; LA64-NEXT:    st.d $a1, $a0, 24
-; LA64-NEXT:    st.d $a4, $a0, 16
-; LA64-NEXT:    st.d $a2, $a0, 8
-; LA64-NEXT:    st.d $a5, $a0, 0
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwod_q_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwod.q.d $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -475,85 +223,13 @@ entry:
 }
 
 define void @vmulwev_q_du(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwev_q_du:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 5
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 4
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr0, 1
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr0, 0
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr1, 5
-; LA32-NEXT:    xvpickve2gr.w $a6, $xr1, 4
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr1, 1
-; LA32-NEXT:    xvpickve2gr.w $t0, $xr1, 0
-; LA32-NEXT:    mulh.wu $t1, $a4, $t0
-; LA32-NEXT:    mul.w $t2, $a3, $t0
-; LA32-NEXT:    add.w $t1, $t2, $t1
-; LA32-NEXT:    sltu $t2, $t1, $t2
-; LA32-NEXT:    mulh.wu $t3, $a3, $t0
-; LA32-NEXT:    add.w $t2, $t3, $t2
-; LA32-NEXT:    mul.w $t3, $a4, $a7
-; LA32-NEXT:    add.w $t1, $t3, $t1
-; LA32-NEXT:    sltu $t3, $t1, $t3
-; LA32-NEXT:    mulh.wu $t4, $a4, $a7
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    add.w $t3, $t2, $t3
-; LA32-NEXT:    mul.w $t4, $a3, $a7
-; LA32-NEXT:    add.w $t5, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t5, $t4
-; LA32-NEXT:    sltu $t2, $t3, $t2
-; LA32-NEXT:    mulh.wu $a3, $a3, $a7
-; LA32-NEXT:    add.w $a3, $a3, $t2
-; LA32-NEXT:    add.w $a3, $a3, $t4
-; LA32-NEXT:    mulh.wu $a7, $a2, $a6
-; LA32-NEXT:    mul.w $t2, $a1, $a6
-; LA32-NEXT:    add.w $a7, $t2, $a7
-; LA32-NEXT:    sltu $t2, $a7, $t2
-; LA32-NEXT:    mulh.wu $t3, $a1, $a6
-; LA32-NEXT:    add.w $t2, $t3, $t2
-; LA32-NEXT:    mul.w $t3, $a2, $a5
-; LA32-NEXT:    add.w $a7, $t3, $a7
-; LA32-NEXT:    sltu $t3, $a7, $t3
-; LA32-NEXT:    mulh.wu $t4, $a2, $a5
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    add.w $t3, $t2, $t3
-; LA32-NEXT:    mul.w $t4, $a1, $a5
-; LA32-NEXT:    add.w $t6, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t6, $t4
-; LA32-NEXT:    sltu $t2, $t3, $t2
-; LA32-NEXT:    mulh.wu $a1, $a1, $a5
-; LA32-NEXT:    add.w $a1, $a1, $t2
-; LA32-NEXT:    add.w $a1, $a1, $t4
-; LA32-NEXT:    mul.w $a4, $a4, $t0
-; LA32-NEXT:    mul.w $a2, $a2, $a6
-; LA32-NEXT:    st.w $a2, $a0, 16
-; LA32-NEXT:    st.w $a4, $a0, 0
-; LA32-NEXT:    st.w $a7, $a0, 20
-; LA32-NEXT:    st.w $t1, $a0, 4
-; LA32-NEXT:    st.w $t6, $a0, 24
-; LA32-NEXT:    st.w $t5, $a0, 8
-; LA32-NEXT:    st.w $a1, $a0, 28
-; LA32-NEXT:    st.w $a3, $a0, 12
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwev_q_du:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 2
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 0
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 2
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 0
-; LA64-NEXT:    mul.d $a5, $a2, $a4
-; LA64-NEXT:    mulh.du $a2, $a2, $a4
-; LA64-NEXT:    mul.d $a4, $a1, $a3
-; LA64-NEXT:    mulh.du $a1, $a1, $a3
-; LA64-NEXT:    st.d $a1, $a0, 24
-; LA64-NEXT:    st.d $a4, $a0, 16
-; LA64-NEXT:    st.d $a2, $a0, 8
-; LA64-NEXT:    st.d $a5, $a0, 0
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwev_q_du:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwev.q.du $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -627,85 +303,13 @@ entry:
 }
 
 define void @vmulwod_q_du(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwod_q_du:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 7
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 6
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr0, 3
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr0, 2
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr1, 7
-; LA32-NEXT:    xvpickve2gr.w $a6, $xr1, 6
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr1, 3
-; LA32-NEXT:    xvpickve2gr.w $t0, $xr1, 2
-; LA32-NEXT:    mulh.wu $t1, $a4, $t0
-; LA32-NEXT:    mul.w $t2, $a3, $t0
-; LA32-NEXT:    add.w $t1, $t2, $t1
-; LA32-NEXT:    sltu $t2, $t1, $t2
-; LA32-NEXT:    mulh.wu $t3, $a3, $t0
-; LA32-NEXT:    add.w $t2, $t3, $t2
-; LA32-NEXT:    mul.w $t3, $a4, $a7
-; LA32-NEXT:    add.w $t1, $t3, $t1
-; LA32-NEXT:    sltu $t3, $t1, $t3
-; LA32-NEXT:    mulh.wu $t4, $a4, $a7
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    add.w $t3, $t2, $t3
-; LA32-NEXT:    mul.w $t4, $a3, $a7
-; LA32-NEXT:    add.w $t5, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t5, $t4
-; LA32-NEXT:    sltu $t2, $t3, $t2
-; LA32-NEXT:    mulh.wu $a3, $a3, $a7
-; LA32-NEXT:    add.w $a3, $a3, $t2
-; LA32-NEXT:    add.w $a3, $a3, $t4
-; LA32-NEXT:    mulh.wu $a7, $a2, $a6
-; LA32-NEXT:    mul.w $t2, $a1, $a6
-; LA32-NEXT:    add.w $a7, $t2, $a7
-; LA32-NEXT:    sltu $t2, $a7, $t2
-; LA32-NEXT:    mulh.wu $t3, $a1, $a6
-; LA32-NEXT:    add.w $t2, $t3, $t2
-; LA32-NEXT:    mul.w $t3, $a2, $a5
-; LA32-NEXT:    add.w $a7, $t3, $a7
-; LA32-NEXT:    sltu $t3, $a7, $t3
-; LA32-NEXT:    mulh.wu $t4, $a2, $a5
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    add.w $t3, $t2, $t3
-; LA32-NEXT:    mul.w $t4, $a1, $a5
-; LA32-NEXT:    add.w $t6, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t6, $t4
-; LA32-NEXT:    sltu $t2, $t3, $t2
-; LA32-NEXT:    mulh.wu $a1, $a1, $a5
-; LA32-NEXT:    add.w $a1, $a1, $t2
-; LA32-NEXT:    add.w $a1, $a1, $t4
-; LA32-NEXT:    mul.w $a4, $a4, $t0
-; LA32-NEXT:    mul.w $a2, $a2, $a6
-; LA32-NEXT:    st.w $a2, $a0, 16
-; LA32-NEXT:    st.w $a4, $a0, 0
-; LA32-NEXT:    st.w $a7, $a0, 20
-; LA32-NEXT:    st.w $t1, $a0, 4
-; LA32-NEXT:    st.w $t6, $a0, 24
-; LA32-NEXT:    st.w $t5, $a0, 8
-; LA32-NEXT:    st.w $a1, $a0, 28
-; LA32-NEXT:    st.w $a3, $a0, 12
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwod_q_du:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 3
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 1
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 3
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 1
-; LA64-NEXT:    mul.d $a5, $a2, $a4
-; LA64-NEXT:    mulh.du $a2, $a2, $a4
-; LA64-NEXT:    mul.d $a4, $a1, $a3
-; LA64-NEXT:    mulh.du $a1, $a1, $a3
-; LA64-NEXT:    st.d $a1, $a0, 24
-; LA64-NEXT:    st.d $a4, $a0, 16
-; LA64-NEXT:    st.d $a2, $a0, 8
-; LA64-NEXT:    st.d $a5, $a0, 0
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwod_q_du:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwod.q.du $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -779,117 +383,13 @@ entry:
 }
 
 define void @vmulwev_q_du_d(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwev_q_du_d:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $sp, $sp, -16
-; LA32-NEXT:    st.w $fp, $sp, 12 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s0, $sp, 8 # 4-byte Folded Spill
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr0, 5
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; LA32-NEXT:    xvpickve2gr.w $a6, $xr0, 1
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 0
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr1, 4
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr1, 0
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr1, 1
-; LA32-NEXT:    xvpickve2gr.w $t0, $xr1, 5
-; LA32-NEXT:    srai.w $t1, $t0, 31
-; LA32-NEXT:    srai.w $t2, $a7, 31
-; LA32-NEXT:    mulh.wu $t3, $a2, $a5
-; LA32-NEXT:    mul.w $t4, $a6, $a5
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t3, $t4
-; LA32-NEXT:    mulh.wu $t5, $a6, $a5
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    mul.w $t5, $a2, $a7
-; LA32-NEXT:    add.w $t3, $t5, $t3
-; LA32-NEXT:    sltu $t5, $t3, $t5
-; LA32-NEXT:    mulh.wu $t6, $a2, $a7
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    add.w $t5, $t4, $t5
-; LA32-NEXT:    mul.w $t6, $a6, $a7
-; LA32-NEXT:    add.w $t7, $t6, $t5
-; LA32-NEXT:    mul.w $t8, $t2, $a2
-; LA32-NEXT:    add.w $fp, $t7, $t8
-; LA32-NEXT:    sltu $s0, $fp, $t7
-; LA32-NEXT:    sltu $t6, $t7, $t6
-; LA32-NEXT:    sltu $t4, $t5, $t4
-; LA32-NEXT:    mulh.wu $a7, $a6, $a7
-; LA32-NEXT:    add.w $a7, $a7, $t4
-; LA32-NEXT:    add.w $a7, $a7, $t6
-; LA32-NEXT:    mul.w $a6, $t2, $a6
-; LA32-NEXT:    mulh.wu $t2, $t2, $a2
-; LA32-NEXT:    add.w $a6, $t2, $a6
-; LA32-NEXT:    add.w $a6, $a6, $t8
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    add.w $a6, $a6, $s0
-; LA32-NEXT:    mulh.wu $a7, $a1, $a4
-; LA32-NEXT:    mul.w $t2, $a3, $a4
-; LA32-NEXT:    add.w $a7, $t2, $a7
-; LA32-NEXT:    sltu $t2, $a7, $t2
-; LA32-NEXT:    mulh.wu $t4, $a3, $a4
-; LA32-NEXT:    add.w $t2, $t4, $t2
-; LA32-NEXT:    mul.w $t4, $a1, $t0
-; LA32-NEXT:    add.w $a7, $t4, $a7
-; LA32-NEXT:    sltu $t4, $a7, $t4
-; LA32-NEXT:    mulh.wu $t5, $a1, $t0
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    add.w $t4, $t2, $t4
-; LA32-NEXT:    mul.w $t5, $a3, $t0
-; LA32-NEXT:    add.w $t6, $t5, $t4
-; LA32-NEXT:    mul.w $t7, $t1, $a1
-; LA32-NEXT:    add.w $t8, $t6, $t7
-; LA32-NEXT:    sltu $s0, $t8, $t6
-; LA32-NEXT:    sltu $t5, $t6, $t5
-; LA32-NEXT:    sltu $t2, $t4, $t2
-; LA32-NEXT:    mulh.wu $t0, $a3, $t0
-; LA32-NEXT:    add.w $t0, $t0, $t2
-; LA32-NEXT:    add.w $t0, $t0, $t5
-; LA32-NEXT:    mul.w $a3, $t1, $a3
-; LA32-NEXT:    mulh.wu $t1, $t1, $a1
-; LA32-NEXT:    add.w $a3, $t1, $a3
-; LA32-NEXT:    add.w $a3, $a3, $t7
-; LA32-NEXT:    add.w $a3, $t0, $a3
-; LA32-NEXT:    add.w $a3, $a3, $s0
-; LA32-NEXT:    mul.w $a2, $a2, $a5
-; LA32-NEXT:    mul.w $a1, $a1, $a4
-; LA32-NEXT:    st.w $a1, $a0, 16
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $a7, $a0, 20
-; LA32-NEXT:    st.w $t3, $a0, 4
-; LA32-NEXT:    st.w $t8, $a0, 24
-; LA32-NEXT:    st.w $fp, $a0, 8
-; LA32-NEXT:    st.w $a3, $a0, 28
-; LA32-NEXT:    st.w $a6, $a0, 12
-; LA32-NEXT:    ld.w $s0, $sp, 8 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $fp, $sp, 12 # 4-byte Folded Reload
-; LA32-NEXT:    addi.w $sp, $sp, 16
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwev_q_du_d:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 2
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 0
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 0
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 2
-; LA64-NEXT:    srai.d $a5, $a4, 63
-; LA64-NEXT:    srai.d $a6, $a3, 63
-; LA64-NEXT:    mulh.du $a7, $a2, $a3
-; LA64-NEXT:    mul.d $a6, $a2, $a6
-; LA64-NEXT:    add.d $a6, $a7, $a6
-; LA64-NEXT:    mulh.du $a7, $a1, $a4
-; LA64-NEXT:    mul.d $a5, $a1, $a5
-; LA64-NEXT:    add.d $a5, $a7, $a5
-; LA64-NEXT:    mul.d $a2, $a2, $a3
-; LA64-NEXT:    mul.d $a1, $a1, $a4
-; LA64-NEXT:    st.d $a1, $a0, 16
-; LA64-NEXT:    st.d $a2, $a0, 0
-; LA64-NEXT:    st.d $a5, $a0, 24
-; LA64-NEXT:    st.d $a6, $a0, 8
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwev_q_du_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwev.q.du.d $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -963,117 +463,13 @@ entry:
 }
 
 define void @vmulwod_q_du_d(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwod_q_du_d:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $sp, $sp, -16
-; LA32-NEXT:    st.w $fp, $sp, 12 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s0, $sp, 8 # 4-byte Folded Spill
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr0, 7
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; LA32-NEXT:    xvpickve2gr.w $a6, $xr0, 3
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 2
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr1, 6
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr1, 2
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr1, 3
-; LA32-NEXT:    xvpickve2gr.w $t0, $xr1, 7
-; LA32-NEXT:    srai.w $t1, $t0, 31
-; LA32-NEXT:    srai.w $t2, $a7, 31
-; LA32-NEXT:    mulh.wu $t3, $a2, $a5
-; LA32-NEXT:    mul.w $t4, $a6, $a5
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t3, $t4
-; LA32-NEXT:    mulh.wu $t5, $a6, $a5
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    mul.w $t5, $a2, $a7
-; LA32-NEXT:    add.w $t3, $t5, $t3
-; LA32-NEXT:    sltu $t5, $t3, $t5
-; LA32-NEXT:    mulh.wu $t6, $a2, $a7
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    add.w $t5, $t4, $t5
-; LA32-NEXT:    mul.w $t6, $a6, $a7
-; LA32-NEXT:    add.w $t7, $t6, $t5
-; LA32-NEXT:    mul.w $t8, $t2, $a2
-; LA32-NEXT:    add.w $fp, $t7, $t8
-; LA32-NEXT:    sltu $s0, $fp, $t7
-; LA32-NEXT:    sltu $t6, $t7, $t6
-; LA32-NEXT:    sltu $t4, $t5, $t4
-; LA32-NEXT:    mulh.wu $a7, $a6, $a7
-; LA32-NEXT:    add.w $a7, $a7, $t4
-; LA32-NEXT:    add.w $a7, $a7, $t6
-; LA32-NEXT:    mul.w $a6, $t2, $a6
-; LA32-NEXT:    mulh.wu $t2, $t2, $a2
-; LA32-NEXT:    add.w $a6, $t2, $a6
-; LA32-NEXT:    add.w $a6, $a6, $t8
-; LA32-NEXT:    add.w $a6, $a7, $a6
-; LA32-NEXT:    add.w $a6, $a6, $s0
-; LA32-NEXT:    mulh.wu $a7, $a1, $a4
-; LA32-NEXT:    mul.w $t2, $a3, $a4
-; LA32-NEXT:    add.w $a7, $t2, $a7
-; LA32-NEXT:    sltu $t2, $a7, $t2
-; LA32-NEXT:    mulh.wu $t4, $a3, $a4
-; LA32-NEXT:    add.w $t2, $t4, $t2
-; LA32-NEXT:    mul.w $t4, $a1, $t0
-; LA32-NEXT:    add.w $a7, $t4, $a7
-; LA32-NEXT:    sltu $t4, $a7, $t4
-; LA32-NEXT:    mulh.wu $t5, $a1, $t0
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    add.w $t4, $t2, $t4
-; LA32-NEXT:    mul.w $t5, $a3, $t0
-; LA32-NEXT:    add.w $t6, $t5, $t4
-; LA32-NEXT:    mul.w $t7, $t1, $a1
-; LA32-NEXT:    add.w $t8, $t6, $t7
-; LA32-NEXT:    sltu $s0, $t8, $t6
-; LA32-NEXT:    sltu $t5, $t6, $t5
-; LA32-NEXT:    sltu $t2, $t4, $t2
-; LA32-NEXT:    mulh.wu $t0, $a3, $t0
-; LA32-NEXT:    add.w $t0, $t0, $t2
-; LA32-NEXT:    add.w $t0, $t0, $t5
-; LA32-NEXT:    mul.w $a3, $t1, $a3
-; LA32-NEXT:    mulh.wu $t1, $t1, $a1
-; LA32-NEXT:    add.w $a3, $t1, $a3
-; LA32-NEXT:    add.w $a3, $a3, $t7
-; LA32-NEXT:    add.w $a3, $t0, $a3
-; LA32-NEXT:    add.w $a3, $a3, $s0
-; LA32-NEXT:    mul.w $a2, $a2, $a5
-; LA32-NEXT:    mul.w $a1, $a1, $a4
-; LA32-NEXT:    st.w $a1, $a0, 16
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $a7, $a0, 20
-; LA32-NEXT:    st.w $t3, $a0, 4
-; LA32-NEXT:    st.w $t8, $a0, 24
-; LA32-NEXT:    st.w $fp, $a0, 8
-; LA32-NEXT:    st.w $a3, $a0, 28
-; LA32-NEXT:    st.w $a6, $a0, 12
-; LA32-NEXT:    ld.w $s0, $sp, 8 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $fp, $sp, 12 # 4-byte Folded Reload
-; LA32-NEXT:    addi.w $sp, $sp, 16
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwod_q_du_d:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 3
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 1
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 1
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 3
-; LA64-NEXT:    srai.d $a5, $a4, 63
-; LA64-NEXT:    srai.d $a6, $a3, 63
-; LA64-NEXT:    mulh.du $a7, $a2, $a3
-; LA64-NEXT:    mul.d $a6, $a2, $a6
-; LA64-NEXT:    add.d $a6, $a7, $a6
-; LA64-NEXT:    mulh.du $a7, $a1, $a4
-; LA64-NEXT:    mul.d $a5, $a1, $a5
-; LA64-NEXT:    add.d $a5, $a7, $a5
-; LA64-NEXT:    mul.d $a2, $a2, $a3
-; LA64-NEXT:    mul.d $a1, $a1, $a4
-; LA64-NEXT:    st.d $a1, $a0, 16
-; LA64-NEXT:    st.d $a2, $a0, 0
-; LA64-NEXT:    st.d $a5, $a0, 24
-; LA64-NEXT:    st.d $a6, $a0, 8
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwod_q_du_d:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwod.q.du.d $xr0, $xr0, $xr1
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -1147,117 +543,13 @@ entry:
 }
 
 define void @vmulwev_q_du_d_1(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwev_q_du_d_1:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $sp, $sp, -16
-; LA32-NEXT:    st.w $fp, $sp, 12 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s0, $sp, 8 # 4-byte Folded Spill
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 4
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 0
-; LA32-NEXT:    xvpickve2gr.w $a6, $xr0, 1
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr0, 5
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr1, 5
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr1, 4
-; LA32-NEXT:    xvpickve2gr.w $t0, $xr1, 1
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr1, 0
-; LA32-NEXT:    srai.w $t1, $a7, 31
-; LA32-NEXT:    srai.w $t2, $a6, 31
-; LA32-NEXT:    mulh.wu $t3, $a2, $a4
-; LA32-NEXT:    mul.w $t4, $a6, $a4
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t3, $t4
-; LA32-NEXT:    mulh.wu $t5, $a6, $a4
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    mul.w $t5, $a2, $t0
-; LA32-NEXT:    add.w $t3, $t5, $t3
-; LA32-NEXT:    sltu $t5, $t3, $t5
-; LA32-NEXT:    mulh.wu $t6, $a2, $t0
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    add.w $t5, $t4, $t5
-; LA32-NEXT:    mul.w $t6, $a6, $t0
-; LA32-NEXT:    add.w $t7, $t6, $t5
-; LA32-NEXT:    mul.w $t8, $a4, $t2
-; LA32-NEXT:    add.w $fp, $t7, $t8
-; LA32-NEXT:    sltu $s0, $fp, $t7
-; LA32-NEXT:    sltu $t6, $t7, $t6
-; LA32-NEXT:    sltu $t4, $t5, $t4
-; LA32-NEXT:    mulh.wu $a6, $a6, $t0
-; LA32-NEXT:    add.w $a6, $a6, $t4
-; LA32-NEXT:    add.w $a6, $a6, $t6
-; LA32-NEXT:    mulh.wu $t4, $a4, $t2
-; LA32-NEXT:    add.w $t4, $t4, $t8
-; LA32-NEXT:    mul.w $t0, $t0, $t2
-; LA32-NEXT:    add.w $t0, $t4, $t0
-; LA32-NEXT:    add.w $a6, $a6, $t0
-; LA32-NEXT:    add.w $a6, $a6, $s0
-; LA32-NEXT:    mulh.wu $t0, $a1, $a3
-; LA32-NEXT:    mul.w $t2, $a7, $a3
-; LA32-NEXT:    add.w $t0, $t2, $t0
-; LA32-NEXT:    sltu $t2, $t0, $t2
-; LA32-NEXT:    mulh.wu $t4, $a7, $a3
-; LA32-NEXT:    add.w $t2, $t4, $t2
-; LA32-NEXT:    mul.w $t4, $a1, $a5
-; LA32-NEXT:    add.w $t0, $t4, $t0
-; LA32-NEXT:    sltu $t4, $t0, $t4
-; LA32-NEXT:    mulh.wu $t5, $a1, $a5
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    add.w $t4, $t2, $t4
-; LA32-NEXT:    mul.w $t5, $a7, $a5
-; LA32-NEXT:    add.w $t6, $t5, $t4
-; LA32-NEXT:    mul.w $t7, $a3, $t1
-; LA32-NEXT:    add.w $t8, $t6, $t7
-; LA32-NEXT:    sltu $s0, $t8, $t6
-; LA32-NEXT:    sltu $t5, $t6, $t5
-; LA32-NEXT:    sltu $t2, $t4, $t2
-; LA32-NEXT:    mulh.wu $a7, $a7, $a5
-; LA32-NEXT:    add.w $a7, $a7, $t2
-; LA32-NEXT:    add.w $a7, $a7, $t5
-; LA32-NEXT:    mulh.wu $t2, $a3, $t1
-; LA32-NEXT:    add.w $t2, $t2, $t7
-; LA32-NEXT:    mul.w $a5, $a5, $t1
-; LA32-NEXT:    add.w $a5, $t2, $a5
-; LA32-NEXT:    add.w $a5, $a7, $a5
-; LA32-NEXT:    add.w $a5, $a5, $s0
-; LA32-NEXT:    mul.w $a2, $a2, $a4
-; LA32-NEXT:    mul.w $a1, $a1, $a3
-; LA32-NEXT:    st.w $a1, $a0, 16
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $t0, $a0, 20
-; LA32-NEXT:    st.w $t3, $a0, 4
-; LA32-NEXT:    st.w $t8, $a0, 24
-; LA32-NEXT:    st.w $fp, $a0, 8
-; LA32-NEXT:    st.w $a5, $a0, 28
-; LA32-NEXT:    st.w $a6, $a0, 12
-; LA32-NEXT:    ld.w $s0, $sp, 8 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $fp, $sp, 12 # 4-byte Folded Reload
-; LA32-NEXT:    addi.w $sp, $sp, 16
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwev_q_du_d_1:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 0
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 2
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 2
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 0
-; LA64-NEXT:    srai.d $a5, $a2, 63
-; LA64-NEXT:    srai.d $a6, $a1, 63
-; LA64-NEXT:    mulh.du $a7, $a1, $a4
-; LA64-NEXT:    mul.d $a6, $a6, $a4
-; LA64-NEXT:    add.d $a6, $a7, $a6
-; LA64-NEXT:    mulh.du $a7, $a2, $a3
-; LA64-NEXT:    mul.d $a5, $a5, $a3
-; LA64-NEXT:    add.d $a5, $a7, $a5
-; LA64-NEXT:    mul.d $a1, $a1, $a4
-; LA64-NEXT:    mul.d $a2, $a2, $a3
-; LA64-NEXT:    st.d $a2, $a0, 16
-; LA64-NEXT:    st.d $a1, $a0, 0
-; LA64-NEXT:    st.d $a5, $a0, 24
-; LA64-NEXT:    st.d $a6, $a0, 8
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwev_q_du_d_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwev.q.du.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b
@@ -1331,117 +623,13 @@ entry:
 }
 
 define void @vmulwod_q_du_d_1(ptr %res, ptr %a, ptr %b) nounwind {
-; LA32-LABEL: vmulwod_q_du_d_1:
-; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $sp, $sp, -16
-; LA32-NEXT:    st.w $fp, $sp, 12 # 4-byte Folded Spill
-; LA32-NEXT:    st.w $s0, $sp, 8 # 4-byte Folded Spill
-; LA32-NEXT:    xvld $xr0, $a1, 0
-; LA32-NEXT:    xvpickve2gr.w $a1, $xr0, 6
-; LA32-NEXT:    xvld $xr1, $a2, 0
-; LA32-NEXT:    xvpickve2gr.w $a2, $xr0, 2
-; LA32-NEXT:    xvpickve2gr.w $a6, $xr0, 3
-; LA32-NEXT:    xvpickve2gr.w $a7, $xr0, 7
-; LA32-NEXT:    xvpickve2gr.w $a5, $xr1, 7
-; LA32-NEXT:    xvpickve2gr.w $a3, $xr1, 6
-; LA32-NEXT:    xvpickve2gr.w $t0, $xr1, 3
-; LA32-NEXT:    xvpickve2gr.w $a4, $xr1, 2
-; LA32-NEXT:    srai.w $t1, $a7, 31
-; LA32-NEXT:    srai.w $t2, $a6, 31
-; LA32-NEXT:    mulh.wu $t3, $a2, $a4
-; LA32-NEXT:    mul.w $t4, $a6, $a4
-; LA32-NEXT:    add.w $t3, $t4, $t3
-; LA32-NEXT:    sltu $t4, $t3, $t4
-; LA32-NEXT:    mulh.wu $t5, $a6, $a4
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    mul.w $t5, $a2, $t0
-; LA32-NEXT:    add.w $t3, $t5, $t3
-; LA32-NEXT:    sltu $t5, $t3, $t5
-; LA32-NEXT:    mulh.wu $t6, $a2, $t0
-; LA32-NEXT:    add.w $t5, $t6, $t5
-; LA32-NEXT:    add.w $t5, $t4, $t5
-; LA32-NEXT:    mul.w $t6, $a6, $t0
-; LA32-NEXT:    add.w $t7, $t6, $t5
-; LA32-NEXT:    mul.w $t8, $a4, $t2
-; LA32-NEXT:    add.w $fp, $t7, $t8
-; LA32-NEXT:    sltu $s0, $fp, $t7
-; LA32-NEXT:    sltu $t6, $t7, $t6
-; LA32-NEXT:    sltu $t4, $t5, $t4
-; LA32-NEXT:    mulh.wu $a6, $a6, $t0
-; LA32-NEXT:    add.w $a6, $a6, $t4
-; LA32-NEXT:    add.w $a6, $a6, $t6
-; LA32-NEXT:    mulh.wu $t4, $a4, $t2
-; LA32-NEXT:    add.w $t4, $t4, $t8
-; LA32-NEXT:    mul.w $t0, $t0, $t2
-; LA32-NEXT:    add.w $t0, $t4, $t0
-; LA32-NEXT:    add.w $a6, $a6, $t0
-; LA32-NEXT:    add.w $a6, $a6, $s0
-; LA32-NEXT:    mulh.wu $t0, $a1, $a3
-; LA32-NEXT:    mul.w $t2, $a7, $a3
-; LA32-NEXT:    add.w $t0, $t2, $t0
-; LA32-NEXT:    sltu $t2, $t0, $t2
-; LA32-NEXT:    mulh.wu $t4, $a7, $a3
-; LA32-NEXT:    add.w $t2, $t4, $t2
-; LA32-NEXT:    mul.w $t4, $a1, $a5
-; LA32-NEXT:    add.w $t0, $t4, $t0
-; LA32-NEXT:    sltu $t4, $t0, $t4
-; LA32-NEXT:    mulh.wu $t5, $a1, $a5
-; LA32-NEXT:    add.w $t4, $t5, $t4
-; LA32-NEXT:    add.w $t4, $t2, $t4
-; LA32-NEXT:    mul.w $t5, $a7, $a5
-; LA32-NEXT:    add.w $t6, $t5, $t4
-; LA32-NEXT:    mul.w $t7, $a3, $t1
-; LA32-NEXT:    add.w $t8, $t6, $t7
-; LA32-NEXT:    sltu $s0, $t8, $t6
-; LA32-NEXT:    sltu $t5, $t6, $t5
-; LA32-NEXT:    sltu $t2, $t4, $t2
-; LA32-NEXT:    mulh.wu $a7, $a7, $a5
-; LA32-NEXT:    add.w $a7, $a7, $t2
-; LA32-NEXT:    add.w $a7, $a7, $t5
-; LA32-NEXT:    mulh.wu $t2, $a3, $t1
-; LA32-NEXT:    add.w $t2, $t2, $t7
-; LA32-NEXT:    mul.w $a5, $a5, $t1
-; LA32-NEXT:    add.w $a5, $t2, $a5
-; LA32-NEXT:    add.w $a5, $a7, $a5
-; LA32-NEXT:    add.w $a5, $a5, $s0
-; LA32-NEXT:    mul.w $a2, $a2, $a4
-; LA32-NEXT:    mul.w $a1, $a1, $a3
-; LA32-NEXT:    st.w $a1, $a0, 16
-; LA32-NEXT:    st.w $a2, $a0, 0
-; LA32-NEXT:    st.w $t0, $a0, 20
-; LA32-NEXT:    st.w $t3, $a0, 4
-; LA32-NEXT:    st.w $t8, $a0, 24
-; LA32-NEXT:    st.w $fp, $a0, 8
-; LA32-NEXT:    st.w $a5, $a0, 28
-; LA32-NEXT:    st.w $a6, $a0, 12
-; LA32-NEXT:    ld.w $s0, $sp, 8 # 4-byte Folded Reload
-; LA32-NEXT:    ld.w $fp, $sp, 12 # 4-byte Folded Reload
-; LA32-NEXT:    addi.w $sp, $sp, 16
-; LA32-NEXT:    ret
-;
-; LA64-LABEL: vmulwod_q_du_d_1:
-; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    xvld $xr0, $a1, 0
-; LA64-NEXT:    xvld $xr1, $a2, 0
-; LA64-NEXT:    xvpickve2gr.d $a1, $xr0, 1
-; LA64-NEXT:    xvpickve2gr.d $a2, $xr0, 3
-; LA64-NEXT:    xvpickve2gr.d $a3, $xr1, 3
-; LA64-NEXT:    xvpickve2gr.d $a4, $xr1, 1
-; LA64-NEXT:    srai.d $a5, $a2, 63
-; LA64-NEXT:    srai.d $a6, $a1, 63
-; LA64-NEXT:    mulh.du $a7, $a1, $a4
-; LA64-NEXT:    mul.d $a6, $a6, $a4
-; LA64-NEXT:    add.d $a6, $a7, $a6
-; LA64-NEXT:    mulh.du $a7, $a2, $a3
-; LA64-NEXT:    mul.d $a5, $a5, $a3
-; LA64-NEXT:    add.d $a5, $a7, $a5
-; LA64-NEXT:    mul.d $a1, $a1, $a4
-; LA64-NEXT:    mul.d $a2, $a2, $a3
-; LA64-NEXT:    st.d $a2, $a0, 16
-; LA64-NEXT:    st.d $a1, $a0, 0
-; LA64-NEXT:    st.d $a5, $a0, 24
-; LA64-NEXT:    st.d $a6, $a0, 8
-; LA64-NEXT:    ret
+; CHECK-LABEL: vmulwod_q_du_d_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    xvld $xr0, $a1, 0
+; CHECK-NEXT:    xvld $xr1, $a2, 0
+; CHECK-NEXT:    xvmulwod.q.du.d $xr0, $xr1, $xr0
+; CHECK-NEXT:    xvst $xr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i64>, ptr %a
   %vb = load <4 x i64>, ptr %b

>From ff07510bfe90339a5eae5c3b833cdef2920cb1f6 Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Fri, 30 Oct 2026 14:52:36 +0800
Subject: [PATCH 4/7] update after rebase

---
 .../LoongArch/lsx/ir-instruction/mulwev_od.ll | 96 +++++++------------
 1 file changed, 32 insertions(+), 64 deletions(-)

diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
index e765898ff58482..ced9b03cc20eb1 100644
--- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
@@ -47,14 +47,10 @@ define void @vmulwev_q_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwev.q.d $vr0, $vr0, $vr1
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_d:
@@ -121,14 +117,10 @@ define void @vmulwod_q_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwod.q.d $vr0, $vr0, $vr1
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_d:
@@ -216,14 +208,10 @@ define void @vmulwev_q_du(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwev.q.du $vr0, $vr0, $vr1
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_du:
@@ -311,14 +299,10 @@ define void @vmulwod_q_du(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwod.q.du $vr0, $vr0, $vr1
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_du:
@@ -385,14 +369,10 @@ define void @vmulwev_q_du_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwev.q.du.d $vr0, $vr0, $vr1
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_du_d:
@@ -459,14 +439,10 @@ define void @vmulwod_q_du_d(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwod.q.du.d $vr0, $vr0, $vr1
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_du_d:
@@ -533,14 +509,10 @@ define void @vmulwev_q_du_d_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwev.q.du.d $vr0, $vr1, $vr0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwev_q_du_d_1:
@@ -607,14 +579,10 @@ define void @vmulwod_q_du_d_1(ptr %res, ptr %a, ptr %b) nounwind {
 ; LA32-NEXT:    vld $vr0, $a1, 0
 ; LA32-NEXT:    vld $vr1, $a2, 0
 ; LA32-NEXT:    vmulwod.q.du.d $vr0, $vr1, $vr0
-; LA32-NEXT:    vpickve2gr.w $a1, $vr0, 0
-; LA32-NEXT:    vpickve2gr.w $a2, $vr0, 1
-; LA32-NEXT:    vpickve2gr.w $a3, $vr0, 2
-; LA32-NEXT:    vpickve2gr.w $a4, $vr0, 3
-; LA32-NEXT:    st.w $a4, $a0, 12
-; LA32-NEXT:    st.w $a3, $a0, 8
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    vstelm.w $vr0, $a0, 12, 3
+; LA32-NEXT:    vstelm.w $vr0, $a0, 8, 2
+; LA32-NEXT:    vstelm.w $vr0, $a0, 4, 1
+; LA32-NEXT:    vstelm.w $vr0, $a0, 0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: vmulwod_q_du_d_1:

>From 28805f45065a4f0e7d3330199e1b93194d39bdb8 Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Fri, 30 Oct 2026 15:15:07 +0800
Subject: [PATCH 5/7] address tangaac's comment

---
 llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 8ab812263fed49..437e541a665fd0 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -9116,7 +9116,7 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
   //   tr = mul ta, tb
   // to:
   //   tr = VMULW{EV/OD}[U/US] t0, t1
-  auto getExtType = [](unsigned Op0, unsigned Op1) -> unsigned {
+  auto getExtType = [](unsigned Op0, unsigned Op1) -> int {
     if (Op0 == ISD::SIGN_EXTEND && Op1 == ISD::SIGN_EXTEND)
       return 0;
     if (Op0 == ISD::ZERO_EXTEND && Op1 == ISD::ZERO_EXTEND)
@@ -9128,7 +9128,7 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
     return -1;
   };
 
-  unsigned ExtType = getExtType(N0.getOpcode(), N1.getOpcode());
+  int ExtType = getExtType(N0.getOpcode(), N1.getOpcode());
   if (ExtType < 0)
     return SDValue();
 

>From 979f4d3ef20756fa381956af6cce5347935a7399 Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Mon, 9 Nov 2026 10:41:45 +0800
Subject: [PATCH 6/7] address heiher's comment

---
 .../LoongArch/LoongArchISelLowering.cpp       |  7 +++-
 .../LoongArch/lsx/ir-instruction/mulwev_od.ll | 40 +++++++++++--------
 2 files changed, 30 insertions(+), 17 deletions(-)

diff --git a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
index 437e541a665fd0..39c1f0de42cd8f 100644
--- a/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
@@ -9132,9 +9132,14 @@ static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
   if (ExtType < 0)
     return SDValue();
 
+  // Avoid the regressions on LA64 introduced by combining
+  // i128-type multiplications.
+  bool IsScalar = (ResTy == MVT::i128);
+  if (IsScalar && ExtType < 2 && Subtarget.is64Bit())
+    return SDValue();
+
   SDValue Src0 = N0.getOperand(0);
   SDValue Src1 = N1.getOperand(0);
-  bool IsScalar = (ResTy == MVT::i128);
   if (IsScalar && (Src0.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
                    Src1.getOpcode() != ISD::EXTRACT_VECTOR_ELT))
     return SDValue();
diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
index ced9b03cc20eb1..7e144fdfa7cbcf 100644
--- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
@@ -55,10 +55,12 @@ define void @vmulwev_q_d(ptr %res, ptr %a, ptr %b) nounwind {
 ;
 ; LA64-LABEL: vmulwev_q_d:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    vld $vr0, $a1, 0
-; LA64-NEXT:    vld $vr1, $a2, 0
-; LA64-NEXT:    vmulwev.q.d $vr0, $vr0, $vr1
-; LA64-NEXT:    vst $vr0, $a0, 0
+; LA64-NEXT:    ld.d $a1, $a1, 0
+; LA64-NEXT:    ld.d $a2, $a2, 0
+; LA64-NEXT:    mul.d $a3, $a1, $a2
+; LA64-NEXT:    mulh.d $a1, $a1, $a2
+; LA64-NEXT:    st.d $a1, $a0, 8
+; LA64-NEXT:    st.d $a3, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -125,10 +127,12 @@ define void @vmulwod_q_d(ptr %res, ptr %a, ptr %b) nounwind {
 ;
 ; LA64-LABEL: vmulwod_q_d:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    vld $vr0, $a1, 0
-; LA64-NEXT:    vld $vr1, $a2, 0
-; LA64-NEXT:    vmulwod.q.d $vr0, $vr0, $vr1
-; LA64-NEXT:    vst $vr0, $a0, 0
+; LA64-NEXT:    ld.d $a1, $a1, 8
+; LA64-NEXT:    ld.d $a2, $a2, 8
+; LA64-NEXT:    mul.d $a3, $a1, $a2
+; LA64-NEXT:    mulh.d $a1, $a1, $a2
+; LA64-NEXT:    st.d $a1, $a0, 8
+; LA64-NEXT:    st.d $a3, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -216,10 +220,12 @@ define void @vmulwev_q_du(ptr %res, ptr %a, ptr %b) nounwind {
 ;
 ; LA64-LABEL: vmulwev_q_du:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    vld $vr0, $a1, 0
-; LA64-NEXT:    vld $vr1, $a2, 0
-; LA64-NEXT:    vmulwev.q.du $vr0, $vr0, $vr1
-; LA64-NEXT:    vst $vr0, $a0, 0
+; LA64-NEXT:    ld.d $a1, $a1, 0
+; LA64-NEXT:    ld.d $a2, $a2, 0
+; LA64-NEXT:    mul.d $a3, $a1, $a2
+; LA64-NEXT:    mulh.du $a1, $a1, $a2
+; LA64-NEXT:    st.d $a1, $a0, 8
+; LA64-NEXT:    st.d $a3, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a
@@ -307,10 +313,12 @@ define void @vmulwod_q_du(ptr %res, ptr %a, ptr %b) nounwind {
 ;
 ; LA64-LABEL: vmulwod_q_du:
 ; LA64:       # %bb.0: # %entry
-; LA64-NEXT:    vld $vr0, $a1, 0
-; LA64-NEXT:    vld $vr1, $a2, 0
-; LA64-NEXT:    vmulwod.q.du $vr0, $vr0, $vr1
-; LA64-NEXT:    vst $vr0, $a0, 0
+; LA64-NEXT:    ld.d $a1, $a1, 8
+; LA64-NEXT:    ld.d $a2, $a2, 8
+; LA64-NEXT:    mul.d $a3, $a1, $a2
+; LA64-NEXT:    mulh.du $a1, $a1, $a2
+; LA64-NEXT:    st.d $a1, $a0, 8
+; LA64-NEXT:    st.d $a3, $a0, 0
 ; LA64-NEXT:    ret
 entry:
   %va = load <2 x i64>, ptr %a

>From accd95de76b529822c119a7157381318d008e611 Mon Sep 17 00:00:00 2001
From: Qi Zhao <zhaoqi01 at loongson.cn>
Date: Mon, 9 Nov 2026 11:06:50 +0800
Subject: [PATCH 7/7] fix tests

---
 .../LoongArch/lsx/ir-instruction/mulwev_od.ll | 126 ++++++++++++++++++
 1 file changed, 126 insertions(+)

diff --git a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
index 7e144fdfa7cbcf..334f93203c397f 100644
--- a/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
+++ b/llvm/test/CodeGen/LoongArch/lsx/ir-instruction/mulwev_od.ll
@@ -3,6 +3,13 @@
 ; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefixes=CHECK,LA64
 
 define void @vmulwev_h_b(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_h_b:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.h.b $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -16,6 +23,13 @@ entry:
 }
 
 define void @vmulwev_w_h(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_w_h:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.w.h $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -29,6 +43,13 @@ entry:
 }
 
 define void @vmulwev_d_w(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_d_w:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.d.w $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -75,6 +96,13 @@ entry:
 }
 
 define void @vmulwod_h_b(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_h_b:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.h.b $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -88,6 +116,13 @@ entry:
 }
 
 define void @vmulwod_w_h(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_w_h:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.w.h $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -101,6 +136,13 @@ entry:
 }
 
 define void @vmulwod_d_w(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_d_w:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.d.w $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -333,6 +375,13 @@ entry:
 }
 
 define void @vmulwev_h_bu_b(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_h_bu_b:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.h.bu.b $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -346,6 +395,13 @@ entry:
 }
 
 define void @vmulwev_w_hu_h(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_w_hu_h:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.w.hu.h $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -359,6 +415,13 @@ entry:
 }
 
 define void @vmulwev_d_wu_w(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_d_wu_w:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.d.wu.w $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -403,6 +466,13 @@ entry:
 }
 
 define void @vmulwod_h_bu_b(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_h_bu_b:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.h.bu.b $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -416,6 +486,13 @@ entry:
 }
 
 define void @vmulwod_w_hu_h(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_w_hu_h:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.w.hu.h $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -429,6 +506,13 @@ entry:
 }
 
 define void @vmulwod_d_wu_w(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_d_wu_w:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.d.wu.w $vr0, $vr0, $vr1
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -473,6 +557,13 @@ entry:
 }
 
 define void @vmulwev_h_bu_b_1(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_h_bu_b_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.h.bu.b $vr0, $vr1, $vr0
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -486,6 +577,13 @@ entry:
 }
 
 define void @vmulwev_w_hu_h_1(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_w_hu_h_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.w.hu.h $vr0, $vr1, $vr0
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -499,6 +597,13 @@ entry:
 }
 
 define void @vmulwev_d_wu_w_1(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwev_d_wu_w_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwev.d.wu.w $vr0, $vr1, $vr0
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b
@@ -543,6 +648,13 @@ entry:
 }
 
 define void @vmulwod_h_bu_b_1(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_h_bu_b_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.h.bu.b $vr0, $vr1, $vr0
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <16 x i8>, ptr %a
   %vb = load <16 x i8>, ptr %b
@@ -556,6 +668,13 @@ entry:
 }
 
 define void @vmulwod_w_hu_h_1(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_w_hu_h_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.w.hu.h $vr0, $vr1, $vr0
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <8 x i16>, ptr %a
   %vb = load <8 x i16>, ptr %b
@@ -569,6 +688,13 @@ entry:
 }
 
 define void @vmulwod_d_wu_w_1(ptr %res, ptr %a, ptr %b) nounwind {
+; CHECK-LABEL: vmulwod_d_wu_w_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vld $vr0, $a1, 0
+; CHECK-NEXT:    vld $vr1, $a2, 0
+; CHECK-NEXT:    vmulwod.d.wu.w $vr0, $vr1, $vr0
+; CHECK-NEXT:    vst $vr0, $a0, 0
+; CHECK-NEXT:    ret
 entry:
   %va = load <4 x i32>, ptr %a
   %vb = load <4 x i32>, ptr %b



More information about the llvm-branch-commits mailing list