[RISCV] Fix handling of nxvXi64 vmsgt(u).vx intrinsics on RV32.

[craig.topper]

Summary

We need to splat the scalar separately and use .vv, but there is
no vmsgt(u).vv. So add isel patterns to select vmslt(u).vv with
swapped operands.

We also need to get VT to use for the splat from an operand rather
than the result since the result VT is nxvXi1.

[HsiangKai]

LGTM.

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

if (!II || !II->SplatOperand)
     return SDValue();
 
   unsigned SplatOp = II->SplatOperand + HasChain;
   assert(SplatOp < Op.getNumOperands());
 
   SmallVector<SDValue, 8> Operands(Op->op_begin(), Op->op_end());
   SDValue &ScalarOp = Operands[SplatOp];
   MVT OpVT = ScalarOp.getSimpleValueType();
-  MVT VT = Op.getSimpleValueType();
   MVT XLenVT = Subtarget.getXLenVT();
 
   // If this isn't a scalar, or its type is XLenVT we're done.
   if (!OpVT.isScalarInteger() || OpVT == XLenVT)
     return SDValue();
 
   // Simplest case is that the operand needs to be promoted to XLenVT.
   if (OpVT.bitsLT(XLenVT)) {
     // If the operand is a constant, sign extend to increase our chances
     // of being able to use a .vi instruction. ANY_EXTEND would become a
     // a zero extend and the simm5 check in isel would fail.
     // FIXME: Should we ignore the upper bits in isel instead?
     unsigned ExtOpc =
         isa<ConstantSDNode>(ScalarOp) ? ISD::SIGN_EXTEND : ISD::ANY_EXTEND;
     ScalarOp = DAG.getNode(ExtOpc, DL, XLenVT, ScalarOp);
     return DAG.getNode(Op->getOpcode(), DL, Op->getVTList(), Operands);
   }
 
+  // Use the previous operand to get the vXi64 VT. The result might be a mask
+  // VT for compares. Using the previous operand assumes that the previous
+  // operand will never have a smaller element size than a scalar operand and
+  // that a widening operation never uses SEW=64.
+  // NOTE: If this fails the below assert, we can probably just find the
+  // element count from any operand or result and use it to construct the VT.
+  assert(II->SplatOperand > 1 && "Unexpected splat operand!");
+  MVT VT = Op.getOperand(SplatOp - 1).getSimpleValueType();
+
   // The more complex case is when the scalar is larger than XLenVT.
   assert(XLenVT == MVT::i32 && OpVT == MVT::i64 &&
          VT.getVectorElementType() == MVT::i64 && "Unexpected VTs!");
 
   // If this is a sign-extended 32-bit constant, we can truncate it and rely
   // on the instruction to sign-extend since SEW>XLEN.
   if (auto *CVal = dyn_cast<ConstantSDNode>(ScalarOp)) {
     if (isInt<32>(CVal->getSExtValue())) {

llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td

 defm "" : VPatBinaryM_VV_VX<"int_riscv_vmsltu", "PseudoVMSLTU", AllIntegerVectors>;
 defm "" : VPatBinaryM_VV_VX<"int_riscv_vmslt", "PseudoVMSLT", AllIntegerVectors>;
 defm "" : VPatBinaryM_VV_VX_VI<"int_riscv_vmsleu", "PseudoVMSLEU", AllIntegerVectors>;
 defm "" : VPatBinaryM_VV_VX_VI<"int_riscv_vmsle", "PseudoVMSLE", AllIntegerVectors>;
 
 defm "" : VPatBinaryM_VX_VI<"int_riscv_vmsgtu", "PseudoVMSGTU", AllIntegerVectors>;
 defm "" : VPatBinaryM_VX_VI<"int_riscv_vmsgt", "PseudoVMSGT", AllIntegerVectors>;
 
+// Match vmsgt with 2 vector operands to vmslt with the operands swapped.
+// Occurs when legalizing vmsgt(u).vx intrinsics for i64 on RV32 since we need
+// to use a more complex splat sequence. Add the pattern for all VTs for
+// consistency.
+foreach vti = AllIntegerVectors in {
+  def : Pat<(vti.Mask (int_riscv_vmsgt (vti.Vector vti.RegClass:$rs2),
+                                       (vti.Vector vti.RegClass:$rs1),
+                                       (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVMSLT_VV_"#vti.LMul.MX) vti.RegClass:$rs1,
+                                                               vti.RegClass:$rs2,
+                                                               GPR:$vl,
+                                                               vti.SEW)>;
+  def : Pat<(vti.Mask (int_riscv_vmsgt_mask (vti.Mask VR:$merge),
+                                            (vti.Vector vti.RegClass:$rs2),
+                                            (vti.Vector vti.RegClass:$rs1),
+                                            (vti.Mask V0),
+                                            (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVMSLT_VV_"#vti.LMul.MX#"_MASK")
+                                                      VR:$merge,
+                                                      vti.RegClass:$rs1,
+                                                      vti.RegClass:$rs2,
+                                                      (vti.Mask V0),
+                                                      GPR:$vl,
+                                                      vti.SEW)>;
+
+  def : Pat<(vti.Mask (int_riscv_vmsgtu (vti.Vector vti.RegClass:$rs2),
+                                        (vti.Vector vti.RegClass:$rs1),
+                                        (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVMSLTU_VV_"#vti.LMul.MX) vti.RegClass:$rs1,
+                                                                vti.RegClass:$rs2,
+                                                                GPR:$vl,
+                                                                vti.SEW)>;
+  def : Pat<(vti.Mask (int_riscv_vmsgtu_mask (vti.Mask VR:$merge),
+                                             (vti.Vector vti.RegClass:$rs2),
+                                             (vti.Vector vti.RegClass:$rs1),
+                                             (vti.Mask V0),
+                                             (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVMSLTU_VV_"#vti.LMul.MX#"_MASK")
+                                                       VR:$merge,
+                                                       vti.RegClass:$rs1,
+                                                       vti.RegClass:$rs2,
+                                                       (vti.Mask V0),
+                                                       GPR:$vl,
+                                                       vti.SEW)>;
+}
+
 // Match vmslt(u).vx intrinsics to vmsle(u).vi if the scalar is -15 to 16. This
 // avoids the user needing to know that there is no vmslt(u).vi instruction.
 // This is limited to vmslt(u).vx as there is no vmsge().vx intrinsic or
 // instruction.
 foreach vti = AllIntegerVectors in {
   def : Pat<(vti.Mask (int_riscv_vmslt (vti.Vector vti.RegClass:$rs1),
-                                       (vti.Scalar simm5_plus1:$rs2), (XLenVT (VLOp GPR:$vl)))),
+                                       (vti.Scalar simm5_plus1:$rs2),
+                                       (XLenVT (VLOp GPR:$vl)))),
             (!cast<Instruction>("PseudoVMSLE_VI_"#vti.LMul.MX) vti.RegClass:$rs1,
                                                                (DecImm simm5_plus1:$rs2),
                                                                GPR:$vl,
                                                                vti.SEW)>;
   def : Pat<(vti.Mask (int_riscv_vmslt_mask (vti.Mask VR:$merge),
                                             (vti.Vector vti.RegClass:$rs1),
                                             (vti.Scalar simm5_plus1:$rs2),
                                             (vti.Mask V0),

llvm/test/CodeGen/RISCV/rvv/vmsgt-rv32.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv32 -mattr=+experimental-v,+f -verify-machineinstrs \
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-v -verify-machineinstrs \
 ; RUN:   --riscv-no-aliases < %s | FileCheck %s
 declare <vscale x 1 x i1> @llvm.riscv.vmsgt.nxv1i8.i8(
   <vscale x 1 x i8>,
   i8,
   i32);
 
 define <vscale x 1 x i1> @intrinsic_vmsgt_vx_nxv1i8_i8(<vscale x 1 x i8> %0, i8 %1, i32 %2) nounwind {
 ; CHECK-LABEL: intrinsic_vmsgt_vx_nxv1i8_i8:
%a = call <vscale x 8 x i1> @llvm.riscv.vmsgt.mask.nxv8i32.i32(
     <vscale x 8 x i32> %1,
     i32 %2,
     <vscale x 8 x i1> %3,
     i32 %4)
 
   ret <vscale x 8 x i1> %a
 }
 
+declare <vscale x 1 x i1> @llvm.riscv.vmsgt.nxv1i64.i64(
+  <vscale x 1 x i64>,
+  i64,
+  i32);
+
+define <vscale x 1 x i1> @intrinsic_vmsgt_vx_nxv1i64_i64(<vscale x 1 x i64> %0, i64 %1, i32 %2) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_vx_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a2, a2, e64,m1,ta,mu
+; CHECK-NEXT:    vmv.v.x v25, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v25, v25, a1
+; CHECK-NEXT:    vmv.v.x v26, a0
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vsrl.vx v26, v26, a1
+; CHECK-NEXT:    vor.vv v25, v26, v25
+; CHECK-NEXT:    vmslt.vv v0, v25, v8
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgt.nxv1i64.i64(
+    <vscale x 1 x i64> %0,
+    i64 %1,
+    i32 %2)
+
+  ret <vscale x 1 x i1> %a
+}
+
+declare <vscale x 1 x i1> @llvm.riscv.vmsgt.mask.nxv1i64.i64(
+  <vscale x 1 x i1>,
+  <vscale x 1 x i64>,
+  i64,
+  <vscale x 1 x i1>,
+  i32);
+
+define <vscale x 1 x i1> @intrinsic_vmsgt_mask_vx_nxv1i64_i64(<vscale x 1 x i1> %0, <vscale x 1 x i64> %1, i64 %2, <vscale x 1 x i1> %3, i32 %4) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_mask_vx_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a3, a2, e64,m1,ta,mu
+; CHECK-NEXT:    vmv.v.x v26, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vmv.v.x v27, a0
+; CHECK-NEXT:    vsll.vx v27, v27, a1
+; CHECK-NEXT:    vsrl.vx v27, v27, a1
+; CHECK-NEXT:    vor.vv v26, v27, v26
+; CHECK-NEXT:    vsetvli a0, a2, e64,m1,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v9
+; CHECK-NEXT:    vmslt.vv v25, v26, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgt.mask.nxv1i64.i64(
+    <vscale x 1 x i1> %0,
+    <vscale x 1 x i64> %1,
+    i64 %2,
+    <vscale x 1 x i1> %3,
+    i32 %4)
+
+  ret <vscale x 1 x i1> %a
+}
+
+declare <vscale x 2 x i1> @llvm.riscv.vmsgt.nxv2i64.i64(
+  <vscale x 2 x i64>,
+  i64,
+  i32);
+
+define <vscale x 2 x i1> @intrinsic_vmsgt_vx_nxv2i64_i64(<vscale x 2 x i64> %0, i64 %1, i32 %2) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_vx_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a2, a2, e64,m2,ta,mu
+; CHECK-NEXT:    vmv.v.x v26, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vmv.v.x v28, a0
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vsrl.vx v28, v28, a1
+; CHECK-NEXT:    vor.vv v26, v28, v26
+; CHECK-NEXT:    vmslt.vv v0, v26, v8
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgt.nxv2i64.i64(
+    <vscale x 2 x i64> %0,
+    i64 %1,
+    i32 %2)
+
+  ret <vscale x 2 x i1> %a
+}
+
+declare <vscale x 2 x i1> @llvm.riscv.vmsgt.mask.nxv2i64.i64(
+  <vscale x 2 x i1>,
+  <vscale x 2 x i64>,
+  i64,
+  <vscale x 2 x i1>,
+  i32);
+
+define <vscale x 2 x i1> @intrinsic_vmsgt_mask_vx_nxv2i64_i64(<vscale x 2 x i1> %0, <vscale x 2 x i64> %1, i64 %2, <vscale x 2 x i1> %3, i32 %4) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_mask_vx_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a3, a2, e64,m2,ta,mu
+; CHECK-NEXT:    vmv.v.x v26, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vmv.v.x v28, a0
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vsrl.vx v28, v28, a1
+; CHECK-NEXT:    vor.vv v26, v28, v26
+; CHECK-NEXT:    vsetvli a0, a2, e64,m2,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v10
+; CHECK-NEXT:    vmslt.vv v25, v26, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgt.mask.nxv2i64.i64(
+    <vscale x 2 x i1> %0,
+    <vscale x 2 x i64> %1,
+    i64 %2,
+    <vscale x 2 x i1> %3,
+    i32 %4)
+
+  ret <vscale x 2 x i1> %a
+}
+
+declare <vscale x 4 x i1> @llvm.riscv.vmsgt.nxv4i64.i64(
+  <vscale x 4 x i64>,
+  i64,
+  i32);
+
+define <vscale x 4 x i1> @intrinsic_vmsgt_vx_nxv4i64_i64(<vscale x 4 x i64> %0, i64 %1, i32 %2) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_vx_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a2, a2, e64,m4,ta,mu
+; CHECK-NEXT:    vmv.v.x v28, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vmv.v.x v12, a0
+; CHECK-NEXT:    vsll.vx v12, v12, a1
+; CHECK-NEXT:    vsrl.vx v12, v12, a1
+; CHECK-NEXT:    vor.vv v28, v12, v28
+; CHECK-NEXT:    vmslt.vv v0, v28, v8
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgt.nxv4i64.i64(
+    <vscale x 4 x i64> %0,
+    i64 %1,
+    i32 %2)
+
+  ret <vscale x 4 x i1> %a
+}
+
+declare <vscale x 4 x i1> @llvm.riscv.vmsgt.mask.nxv4i64.i64(
+  <vscale x 4 x i1>,
+  <vscale x 4 x i64>,
+  i64,
+  <vscale x 4 x i1>,
+  i32);
+
+define <vscale x 4 x i1> @intrinsic_vmsgt_mask_vx_nxv4i64_i64(<vscale x 4 x i1> %0, <vscale x 4 x i64> %1, i64 %2, <vscale x 4 x i1> %3, i32 %4) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_mask_vx_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a3, a2, e64,m4,ta,mu
+; CHECK-NEXT:    vmv.v.x v28, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vmv.v.x v16, a0
+; CHECK-NEXT:    vsll.vx v16, v16, a1
+; CHECK-NEXT:    vsrl.vx v16, v16, a1
+; CHECK-NEXT:    vor.vv v28, v16, v28
+; CHECK-NEXT:    vsetvli a0, a2, e64,m4,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v12
+; CHECK-NEXT:    vmslt.vv v25, v28, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgt.mask.nxv4i64.i64(
+    <vscale x 4 x i1> %0,
+    <vscale x 4 x i64> %1,
+    i64 %2,
+    <vscale x 4 x i1> %3,
+    i32 %4)
+
+  ret <vscale x 4 x i1> %a
+}
+
 define <vscale x 1 x i1> @intrinsic_vmsgt_vi_nxv1i8_i8(<vscale x 1 x i8> %0, i32 %1) nounwind {
 ; CHECK-LABEL: intrinsic_vmsgt_vi_nxv1i8_i8:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vsetvli a0, a0, e8,mf8,ta,mu
 ; CHECK-NEXT:    vmsgt.vi v0, v8, 9
 ; CHECK-NEXT:    jalr zero, 0(ra)
 entry:
   %a = call <vscale x 1 x i1> @llvm.riscv.vmsgt.nxv1i8.i8(
%a = call <vscale x 8 x i1> @llvm.riscv.vmsgt.mask.nxv8i32.i32(
     <vscale x 8 x i1> %0,
     <vscale x 8 x i32> %1,
     i32 9,
     <vscale x 8 x i1> %2,
     i32 %3)
 
   ret <vscale x 8 x i1> %a
 }
+
+define <vscale x 1 x i1> @intrinsic_vmsgt_vi_nxv1i64_i64(<vscale x 1 x i64> %0, i32 %1) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_vi_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a0, a0, e64,m1,ta,mu
+; CHECK-NEXT:    vmsgt.vi v0, v8, 9
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgt.nxv1i64.i64(
+    <vscale x 1 x i64> %0,
+    i64 9,
+    i32 %1)
+
+  ret <vscale x 1 x i1> %a
+}
+
+define <vscale x 1 x i1> @intrinsic_vmsgt_mask_vi_nxv1i64_i64(<vscale x 1 x i1> %0, <vscale x 1 x i64> %1, <vscale x 1 x i1> %2, i32 %3) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_mask_vi_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a0, a0, e64,m1,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v9
+; CHECK-NEXT:    vmsgt.vi v25, v8, 9, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgt.mask.nxv1i64.i64(
+    <vscale x 1 x i1> %0,
+    <vscale x 1 x i64> %1,
+    i64 9,
+    <vscale x 1 x i1> %2,
+    i32 %3)
+
+  ret <vscale x 1 x i1> %a
+}
+
+define <vscale x 2 x i1> @intrinsic_vmsgt_vi_nxv2i64_i64(<vscale x 2 x i64> %0, i32 %1) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_vi_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a0, a0, e64,m2,ta,mu
+; CHECK-NEXT:    vmsgt.vi v0, v8, 9
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgt.nxv2i64.i64(
+    <vscale x 2 x i64> %0,
+    i64 9,
+    i32 %1)
+
+  ret <vscale x 2 x i1> %a
+}
+
+define <vscale x 2 x i1> @intrinsic_vmsgt_mask_vi_nxv2i64_i64(<vscale x 2 x i1> %0, <vscale x 2 x i64> %1, <vscale x 2 x i1> %2, i32 %3) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_mask_vi_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a0, a0, e64,m2,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v10
+; CHECK-NEXT:    vmsgt.vi v25, v8, 9, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgt.mask.nxv2i64.i64(
+    <vscale x 2 x i1> %0,
+    <vscale x 2 x i64> %1,
+    i64 9,
+    <vscale x 2 x i1> %2,
+    i32 %3)
+
+  ret <vscale x 2 x i1> %a
+}
+
+define <vscale x 4 x i1> @intrinsic_vmsgt_vi_nxv4i64_i64(<vscale x 4 x i64> %0, i32 %1) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_vi_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a0, a0, e64,m4,ta,mu
+; CHECK-NEXT:    vmsgt.vi v0, v8, 9
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgt.nxv4i64.i64(
+    <vscale x 4 x i64> %0,
+    i64 9,
+    i32 %1)
+
+  ret <vscale x 4 x i1> %a
+}
+
+define <vscale x 4 x i1> @intrinsic_vmsgt_mask_vi_nxv4i64_i64(<vscale x 4 x i1> %0, <vscale x 4 x i64> %1, <vscale x 4 x i1> %2, i32 %3) nounwind {
+; CHECK-LABEL: intrinsic_vmsgt_mask_vi_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a0, a0, e64,m4,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v12
+; CHECK-NEXT:    vmsgt.vi v25, v8, 9, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgt.mask.nxv4i64.i64(
+    <vscale x 4 x i1> %0,
+    <vscale x 4 x i64> %1,
+    i64 9,
+    <vscale x 4 x i1> %2,
+    i32 %3)
+
+  ret <vscale x 4 x i1> %a
+}

llvm/test/CodeGen/RISCV/rvv/vmsgt-rv64.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv64 -mattr=+experimental-v,+d -verify-machineinstrs \
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs \
 ; RUN:   --riscv-no-aliases < %s | FileCheck %s
 declare <vscale x 1 x i1> @llvm.riscv.vmsgt.nxv1i8.i8(
   <vscale x 1 x i8>,
   i8,
   i64);
 
 define <vscale x 1 x i1> @intrinsic_vmsgt_vx_nxv1i8_i8(<vscale x 1 x i8> %0, i8 %1, i64 %2) nounwind {
 ; CHECK-LABEL: intrinsic_vmsgt_vx_nxv1i8_i8:

llvm/test/CodeGen/RISCV/rvv/vmsgtu-rv32.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv32 -mattr=+experimental-v,+f -verify-machineinstrs \
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-v -verify-machineinstrs \
 ; RUN:   --riscv-no-aliases < %s | FileCheck %s
 declare <vscale x 1 x i1> @llvm.riscv.vmsgtu.nxv1i8.i8(
   <vscale x 1 x i8>,
   i8,
   i32);
 
 define <vscale x 1 x i1> @intrinsic_vmsgtu_vx_nxv1i8_i8(<vscale x 1 x i8> %0, i8 %1, i32 %2) nounwind {
 ; CHECK-LABEL: intrinsic_vmsgtu_vx_nxv1i8_i8:
%a = call <vscale x 8 x i1> @llvm.riscv.vmsgtu.mask.nxv8i32.i32(
     <vscale x 8 x i32> %1,
     i32 %2,
     <vscale x 8 x i1> %3,
     i32 %4)
 
   ret <vscale x 8 x i1> %a
 }
 
+declare <vscale x 1 x i1> @llvm.riscv.vmsgtu.nxv1i64.i64(
+  <vscale x 1 x i64>,
+  i64,
+  i32);
+
+define <vscale x 1 x i1> @intrinsic_vmsgtu_vx_nxv1i64_i64(<vscale x 1 x i64> %0, i64 %1, i32 %2) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_vx_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a2, a2, e64,m1,ta,mu
+; CHECK-NEXT:    vmv.v.x v25, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v25, v25, a1
+; CHECK-NEXT:    vmv.v.x v26, a0
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vsrl.vx v26, v26, a1
+; CHECK-NEXT:    vor.vv v25, v26, v25
+; CHECK-NEXT:    vmsltu.vv v0, v25, v8
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgtu.nxv1i64.i64(
+    <vscale x 1 x i64> %0,
+    i64 %1,
+    i32 %2)
+
+  ret <vscale x 1 x i1> %a
+}
+
+declare <vscale x 1 x i1> @llvm.riscv.vmsgtu.mask.nxv1i64.i64(
+  <vscale x 1 x i1>,
+  <vscale x 1 x i64>,
+  i64,
+  <vscale x 1 x i1>,
+  i32);
+
+define <vscale x 1 x i1> @intrinsic_vmsgtu_mask_vx_nxv1i64_i64(<vscale x 1 x i1> %0, <vscale x 1 x i64> %1, i64 %2, <vscale x 1 x i1> %3, i32 %4) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_mask_vx_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a3, a2, e64,m1,ta,mu
+; CHECK-NEXT:    vmv.v.x v26, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vmv.v.x v27, a0
+; CHECK-NEXT:    vsll.vx v27, v27, a1
+; CHECK-NEXT:    vsrl.vx v27, v27, a1
+; CHECK-NEXT:    vor.vv v26, v27, v26
+; CHECK-NEXT:    vsetvli a0, a2, e64,m1,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v9
+; CHECK-NEXT:    vmsltu.vv v25, v26, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgtu.mask.nxv1i64.i64(
+    <vscale x 1 x i1> %0,
+    <vscale x 1 x i64> %1,
+    i64 %2,
+    <vscale x 1 x i1> %3,
+    i32 %4)
+
+  ret <vscale x 1 x i1> %a
+}
+
+declare <vscale x 2 x i1> @llvm.riscv.vmsgtu.nxv2i64.i64(
+  <vscale x 2 x i64>,
+  i64,
+  i32);
+
+define <vscale x 2 x i1> @intrinsic_vmsgtu_vx_nxv2i64_i64(<vscale x 2 x i64> %0, i64 %1, i32 %2) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_vx_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a2, a2, e64,m2,ta,mu
+; CHECK-NEXT:    vmv.v.x v26, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vmv.v.x v28, a0
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vsrl.vx v28, v28, a1
+; CHECK-NEXT:    vor.vv v26, v28, v26
+; CHECK-NEXT:    vmsltu.vv v0, v26, v8
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgtu.nxv2i64.i64(
+    <vscale x 2 x i64> %0,
+    i64 %1,
+    i32 %2)
+
+  ret <vscale x 2 x i1> %a
+}
+
+declare <vscale x 2 x i1> @llvm.riscv.vmsgtu.mask.nxv2i64.i64(
+  <vscale x 2 x i1>,
+  <vscale x 2 x i64>,
+  i64,
+  <vscale x 2 x i1>,
+  i32);
+
+define <vscale x 2 x i1> @intrinsic_vmsgtu_mask_vx_nxv2i64_i64(<vscale x 2 x i1> %0, <vscale x 2 x i64> %1, i64 %2, <vscale x 2 x i1> %3, i32 %4) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_mask_vx_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a3, a2, e64,m2,ta,mu
+; CHECK-NEXT:    vmv.v.x v26, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v26, v26, a1
+; CHECK-NEXT:    vmv.v.x v28, a0
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vsrl.vx v28, v28, a1
+; CHECK-NEXT:    vor.vv v26, v28, v26
+; CHECK-NEXT:    vsetvli a0, a2, e64,m2,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v10
+; CHECK-NEXT:    vmsltu.vv v25, v26, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgtu.mask.nxv2i64.i64(
+    <vscale x 2 x i1> %0,
+    <vscale x 2 x i64> %1,
+    i64 %2,
+    <vscale x 2 x i1> %3,
+    i32 %4)
+
+  ret <vscale x 2 x i1> %a
+}
+
+declare <vscale x 4 x i1> @llvm.riscv.vmsgtu.nxv4i64.i64(
+  <vscale x 4 x i64>,
+  i64,
+  i32);
+
+define <vscale x 4 x i1> @intrinsic_vmsgtu_vx_nxv4i64_i64(<vscale x 4 x i64> %0, i64 %1, i32 %2) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_vx_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a2, a2, e64,m4,ta,mu
+; CHECK-NEXT:    vmv.v.x v28, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vmv.v.x v12, a0
+; CHECK-NEXT:    vsll.vx v12, v12, a1
+; CHECK-NEXT:    vsrl.vx v12, v12, a1
+; CHECK-NEXT:    vor.vv v28, v12, v28
+; CHECK-NEXT:    vmsltu.vv v0, v28, v8
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgtu.nxv4i64.i64(
+    <vscale x 4 x i64> %0,
+    i64 %1,
+    i32 %2)
+
+  ret <vscale x 4 x i1> %a
+}
+
+declare <vscale x 4 x i1> @llvm.riscv.vmsgtu.mask.nxv4i64.i64(
+  <vscale x 4 x i1>,
+  <vscale x 4 x i64>,
+  i64,
+  <vscale x 4 x i1>,
+  i32);
+
+define <vscale x 4 x i1> @intrinsic_vmsgtu_mask_vx_nxv4i64_i64(<vscale x 4 x i1> %0, <vscale x 4 x i64> %1, i64 %2, <vscale x 4 x i1> %3, i32 %4) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_mask_vx_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a3, a2, e64,m4,ta,mu
+; CHECK-NEXT:    vmv.v.x v28, a1
+; CHECK-NEXT:    addi a1, zero, 32
+; CHECK-NEXT:    vsll.vx v28, v28, a1
+; CHECK-NEXT:    vmv.v.x v16, a0
+; CHECK-NEXT:    vsll.vx v16, v16, a1
+; CHECK-NEXT:    vsrl.vx v16, v16, a1
+; CHECK-NEXT:    vor.vv v28, v16, v28
+; CHECK-NEXT:    vsetvli a0, a2, e64,m4,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v12
+; CHECK-NEXT:    vmsltu.vv v25, v28, v8, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgtu.mask.nxv4i64.i64(
+    <vscale x 4 x i1> %0,
+    <vscale x 4 x i64> %1,
+    i64 %2,
+    <vscale x 4 x i1> %3,
+    i32 %4)
+
+  ret <vscale x 4 x i1> %a
+}
+
 define <vscale x 1 x i1> @intrinsic_vmsgtu_vi_nxv1i8_i8(<vscale x 1 x i8> %0, i32 %1) nounwind {
 ; CHECK-LABEL: intrinsic_vmsgtu_vi_nxv1i8_i8:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    vsetvli a0, a0, e8,mf8,ta,mu
 ; CHECK-NEXT:    vmsgtu.vi v0, v8, 9
 ; CHECK-NEXT:    jalr zero, 0(ra)
 entry:
   %a = call <vscale x 1 x i1> @llvm.riscv.vmsgtu.nxv1i8.i8(
%a = call <vscale x 8 x i1> @llvm.riscv.vmsgtu.mask.nxv8i32.i32(
     <vscale x 8 x i1> %0,
     <vscale x 8 x i32> %1,
     i32 9,
     <vscale x 8 x i1> %2,
     i32 %3)
 
   ret <vscale x 8 x i1> %a
 }
+
+define <vscale x 1 x i1> @intrinsic_vmsgtu_vi_nxv1i64_i64(<vscale x 1 x i64> %0, i32 %1) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_vi_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a0, a0, e64,m1,ta,mu
+; CHECK-NEXT:    vmsgtu.vi v0, v8, 9
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgtu.nxv1i64.i64(
+    <vscale x 1 x i64> %0,
+    i64 9,
+    i32 %1)
+
+  ret <vscale x 1 x i1> %a
+}
+
+define <vscale x 1 x i1> @intrinsic_vmsgtu_mask_vi_nxv1i64_i64(<vscale x 1 x i1> %0, <vscale x 1 x i64> %1, <vscale x 1 x i1> %2, i32 %3) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_mask_vi_nxv1i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a0, a0, e64,m1,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v9
+; CHECK-NEXT:    vmsgtu.vi v25, v8, 9, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 1 x i1> @llvm.riscv.vmsgtu.mask.nxv1i64.i64(
+    <vscale x 1 x i1> %0,
+    <vscale x 1 x i64> %1,
+    i64 9,
+    <vscale x 1 x i1> %2,
+    i32 %3)
+
+  ret <vscale x 1 x i1> %a
+}
+
+define <vscale x 2 x i1> @intrinsic_vmsgtu_vi_nxv2i64_i64(<vscale x 2 x i64> %0, i32 %1) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_vi_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a0, a0, e64,m2,ta,mu
+; CHECK-NEXT:    vmsgtu.vi v0, v8, 9
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgtu.nxv2i64.i64(
+    <vscale x 2 x i64> %0,
+    i64 9,
+    i32 %1)
+
+  ret <vscale x 2 x i1> %a
+}
+
+define <vscale x 2 x i1> @intrinsic_vmsgtu_mask_vi_nxv2i64_i64(<vscale x 2 x i1> %0, <vscale x 2 x i64> %1, <vscale x 2 x i1> %2, i32 %3) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_mask_vi_nxv2i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a0, a0, e64,m2,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v10
+; CHECK-NEXT:    vmsgtu.vi v25, v8, 9, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 2 x i1> @llvm.riscv.vmsgtu.mask.nxv2i64.i64(
+    <vscale x 2 x i1> %0,
+    <vscale x 2 x i64> %1,
+    i64 9,
+    <vscale x 2 x i1> %2,
+    i32 %3)
+
+  ret <vscale x 2 x i1> %a
+}
+
+define <vscale x 4 x i1> @intrinsic_vmsgtu_vi_nxv4i64_i64(<vscale x 4 x i64> %0, i32 %1) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_vi_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsetvli a0, a0, e64,m4,ta,mu
+; CHECK-NEXT:    vmsgtu.vi v0, v8, 9
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgtu.nxv4i64.i64(
+    <vscale x 4 x i64> %0,
+    i64 9,
+    i32 %1)
+
+  ret <vscale x 4 x i1> %a
+}
+
+define <vscale x 4 x i1> @intrinsic_vmsgtu_mask_vi_nxv4i64_i64(<vscale x 4 x i1> %0, <vscale x 4 x i64> %1, <vscale x 4 x i1> %2, i32 %3) nounwind {
+; CHECK-LABEL: intrinsic_vmsgtu_mask_vi_nxv4i64_i64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmv1r.v v25, v0
+; CHECK-NEXT:    vsetvli a0, a0, e64,m4,tu,mu
+; CHECK-NEXT:    vmv1r.v v0, v12
+; CHECK-NEXT:    vmsgtu.vi v25, v8, 9, v0.t
+; CHECK-NEXT:    vmv1r.v v0, v25
+; CHECK-NEXT:    jalr zero, 0(ra)
+entry:
+  %a = call <vscale x 4 x i1> @llvm.riscv.vmsgtu.mask.nxv4i64.i64(
+    <vscale x 4 x i1> %0,
+    <vscale x 4 x i64> %1,
+    i64 9,
+    <vscale x 4 x i1> %2,
+    i32 %3)
+
+  ret <vscale x 4 x i1> %a
+}

llvm/test/CodeGen/RISCV/rvv/vmsgtu-rv64.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=riscv64 -mattr=+experimental-v,+d -verify-machineinstrs \
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs \
 ; RUN:   --riscv-no-aliases < %s | FileCheck %s
 declare <vscale x 1 x i1> @llvm.riscv.vmsgtu.nxv1i8.i8(
   <vscale x 1 x i8>,
   i8,
   i64);
 
 define <vscale x 1 x i1> @intrinsic_vmsgtu_vx_nxv1i8_i8(<vscale x 1 x i8> %0, i8 %1, i64 %2) nounwind {
 ; CHECK-LABEL: intrinsic_vmsgtu_vx_nxv1i8_i8: