[RISCV] Add support for selecting vrgather.vx/vi for fixed vector splat shuffles.

[craig.topper]

Summary

The test cases extract a fixed element from a vector and splat it
into a vector. This gets DAG combined into a splat shuffle.

I've used some very wide vectors in the test to make sure we have
at least a couple tests where the element doesn't fit into the
uimm5 immediate of vrgather.vi so we fall back to vrgather.vx.

[craig.topper]

Rebase

[craig.topper]

clang-format

[frasercrmck]

LGTM

llvm/lib/Target/RISCV/RISCVISelLowering.h

 //===-- RISCVISelLowering.h - RISCV DAG Lowering Interface ------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
enum NodeType : unsigned {
   FDIV_VL,
   FNEG_VL,
   FMA_VL,
 
   // Set mask vector to all zeros or ones.
   VMCLR_VL,
   VMSET_VL,
 
+  // Matches the semantics of vrgather.vx with an extra operand for VL.
+  VRGATHER_VX_VL,
+
   // Memory opcodes start here.
   VLE_VL = ISD::FIRST_TARGET_MEMORY_OPCODE,
   VSE_VL,
 
   // WARNING: Do not add anything in the end unless you want the node to
   // have memop! In fact, starting from FIRST_TARGET_MEMORY_OPCODE all
   // opcodes will be thought as target memory ops!
 };

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

 //===-- RISCVISelLowering.cpp - RISCV DAG Lowering Implementation  --------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
if (Subtarget.useRVVForFixedLengthVectors()) {
         // By default everything must be expanded.
         for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op)
           setOperationAction(Op, VT, Expand);
 
         // We use EXTRACT_SUBVECTOR as a "cast" from scalable to fixed.
         setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal);
 
         setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
+        setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
 
         setOperationAction(ISD::LOAD, VT, Custom);
         setOperationAction(ISD::STORE, VT, Custom);
         setOperationAction(ISD::ADD, VT, Custom);
         setOperationAction(ISD::MUL, VT, Custom);
         setOperationAction(ISD::SUB, VT, Custom);
         setOperationAction(ISD::AND, VT, Custom);
         setOperationAction(ISD::OR, VT, Custom);
if (Subtarget.useRVVForFixedLengthVectors()) {
         // By default everything must be expanded.
         for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op)
           setOperationAction(Op, VT, Expand);
 
         // We use EXTRACT_SUBVECTOR as a "cast" from scalable to fixed.
         setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal);
 
         setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
+        setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
 
         setOperationAction(ISD::LOAD, VT, Custom);
         setOperationAction(ISD::STORE, VT, Custom);
         setOperationAction(ISD::FADD, VT, Custom);
         setOperationAction(ISD::FSUB, VT, Custom);
         setOperationAction(ISD::FMUL, VT, Custom);
         setOperationAction(ISD::FDIV, VT, Custom);
         setOperationAction(ISD::FNEG, VT, Custom);
unsigned Opc = VT.isFloatingPoint() ? RISCVISD::VFMV_V_F_VL
                                         : RISCVISD::VMV_V_X_VL;
     Splat = DAG.getNode(Opc, DL, ContainerVT, Splat, VL);
     return convertFromScalableVector(VT, Splat, DAG, Subtarget);
   }
 
   return SDValue();
 }
 
+static SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG,
+                                   const RISCVSubtarget &Subtarget) {
+  SDValue V1 = Op.getOperand(0);
+  SDLoc DL(Op);
+  MVT VT = Op.getSimpleValueType();
+  ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode());
+
+  if (SVN->isSplat()) {
+    int Lane = SVN->getSplatIndex();
+    if (Lane >= 0) {
+      MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
+
+      V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
+      assert(Lane < (int)VT.getVectorNumElements() && "Unexpected lane!");
+
+      MVT XLenVT = Subtarget.getXLenVT();
+      SDValue VL = DAG.getConstant(VT.getVectorNumElements(), DL, XLenVT);
+      MVT MaskVT =
+          MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
+      SDValue Mask = DAG.getNode(RISCVISD::VMSET_VL, DL, MaskVT, VL);
+      SDValue Gather =
+          DAG.getNode(RISCVISD::VRGATHER_VX_VL, DL, ContainerVT, V1,
+                      DAG.getConstant(Lane, DL, XLenVT), Mask, VL);
+      return convertFromScalableVector(VT, Gather, DAG, Subtarget);
+    }
+  }
+
+  return SDValue();
+}
+
 SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
                                             SelectionDAG &DAG) const {
   switch (Op.getOpcode()) {
   default:
     report_fatal_error("unimplemented operand");
   case ISD::GlobalAddress:
     return lowerGlobalAddress(Op, DAG);
   case ISD::BlockAddress:
SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
   case ISD::VECREDUCE_OR:
   case ISD::VECREDUCE_XOR:
     return lowerVECREDUCE(Op, DAG);
   case ISD::VECREDUCE_FADD:
   case ISD::VECREDUCE_SEQ_FADD:
     return lowerFPVECREDUCE(Op, DAG);
   case ISD::BUILD_VECTOR:
     return lowerBUILD_VECTOR(Op, DAG, Subtarget);
+  case ISD::VECTOR_SHUFFLE:
+    return lowerVECTOR_SHUFFLE(Op, DAG, Subtarget);
   case ISD::LOAD:
     return lowerFixedLengthVectorLoadToRVV(Op, DAG);
   case ISD::STORE:
     return lowerFixedLengthVectorStoreToRVV(Op, DAG);
   case ISD::ADD:
     return lowerToScalableOp(Op, DAG, RISCVISD::ADD_VL);
   case ISD::SUB:
     return lowerToScalableOp(Op, DAG, RISCVISD::SUB_VL);
#define NODE_NAME_CASE(NODE)                                                   \
   NODE_NAME_CASE(FADD_VL)
   NODE_NAME_CASE(FSUB_VL)
   NODE_NAME_CASE(FMUL_VL)
   NODE_NAME_CASE(FDIV_VL)
   NODE_NAME_CASE(FNEG_VL)
   NODE_NAME_CASE(FMA_VL)
   NODE_NAME_CASE(VMCLR_VL)
   NODE_NAME_CASE(VMSET_VL)
+  NODE_NAME_CASE(VRGATHER_VX_VL)
   NODE_NAME_CASE(VLE_VL)
   NODE_NAME_CASE(VSE_VL)
   }
   // clang-format on
   return nullptr;
 #undef NODE_NAME_CASE
 }
 

llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td

def SDT_RISCVVecFMA_VL : SDTypeProfile<1, 5, [SDTCisSameAs<0, 1>,
                                               SDTCisSameAs<0, 2>,
                                               SDTCisSameAs<0, 3>,
                                               SDTCisVec<0>, SDTCisFP<0>,
                                               SDTCVecEltisVT<4, i1>,
                                               SDTCisSameNumEltsAs<0, 4>,
                                               SDTCisVT<5, XLenVT>]>;
 def riscv_fma_vl : SDNode<"RISCVISD::FMA_VL", SDT_RISCVVecFMA_VL>;
 
+def riscv_vrgather_vx_vl : SDNode<"RISCVISD::VRGATHER_VX_VL",
+                                  SDTypeProfile<1, 4, [SDTCisVec<0>,
+                                                       SDTCisSameAs<0, 1>,
+                                                       SDTCisVT<2, XLenVT>,
+                                                       SDTCVecEltisVT<3, i1>,
+                                                       SDTCisSameNumEltsAs<0, 3>,
+                                                       SDTCisVT<4, XLenVT>]>>;
+
 def SDT_RISCVVMSETCLR_VL : SDTypeProfile<1, 1, [SDTCisVec<0>,
                                                 SDTCVecEltisVT<0, i1>,
                                                 SDTCisVT<1, XLenVT>]>;
 def riscv_vmclr_vl : SDNode<"RISCVISD::VMCLR_VL", SDT_RISCVVMSETCLR_VL>;
 def riscv_vmset_vl : SDNode<"RISCVISD::VMSET_VL", SDT_RISCVVMSETCLR_VL>;
 
 def true_mask           : PatLeaf<(riscv_vmset_vl (XLenVT srcvalue))>;
 

 foreach vti = AllFloatVectors in {
   def : Pat<(riscv_fneg_vl (vti.Vector vti.RegClass:$rs), (vti.Mask true_mask),
                            (XLenVT (VLOp GPR:$vl))),
             (!cast<Instruction>("PseudoVFSGNJN_VV_"# vti.LMul.MX)
                  vti.RegClass:$rs, vti.RegClass:$rs, GPR:$vl, vti.SEW)>;
 }
 
 } // Predicates = [HasStdExtV, HasStdExtF]
+
+// 17.4. Vector Register GAther Instruction
+let Predicates = [HasStdExtV] in {
+
+foreach vti = AllIntegerVectors in {
+  def : Pat<(vti.Vector (riscv_vrgather_vx_vl vti.RegClass:$rs2, GPR:$rs1,
+                                              (vti.Mask true_mask),
+                                              (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVRGATHER_VX_"# vti.LMul.MX)
+                 vti.RegClass:$rs2, GPR:$rs1, GPR:$vl, vti.SEW)>;
+  def : Pat<(vti.Vector (riscv_vrgather_vx_vl vti.RegClass:$rs2, uimm5:$imm,
+                                              (vti.Mask true_mask),
+                                              (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVRGATHER_VI_"# vti.LMul.MX)
+                 vti.RegClass:$rs2, uimm5:$imm, GPR:$vl, vti.SEW)>;
+}
+
+} // Predicates = [HasStdExtV]
+
+let Predicates = [HasStdExtV, HasStdExtF] in {
+
+foreach vti = AllFloatVectors in {
+  def : Pat<(vti.Vector (riscv_vrgather_vx_vl vti.RegClass:$rs2, GPR:$rs1,
+                                              (vti.Mask true_mask),
+                                              (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVRGATHER_VX_"# vti.LMul.MX)
+                 vti.RegClass:$rs2, GPR:$rs1, GPR:$vl, vti.SEW)>;
+  def : Pat<(vti.Vector (riscv_vrgather_vx_vl vti.RegClass:$rs2, uimm5:$imm,
+                                              (vti.Mask true_mask),
+                                              (XLenVT (VLOp GPR:$vl)))),
+            (!cast<Instruction>("PseudoVRGATHER_VI_"# vti.LMul.MX)
+                 vti.RegClass:$rs2, uimm5:$imm, GPR:$vl, vti.SEW)>;
+}
+
+} // Predicates = [HasStdExtV, HasStdExtF]

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-fp-vrgather.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -target-abi=ilp32d -mattr=+experimental-v,+experimental-zfh,+f,+d -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=8 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX8
+; RUN: llc -mtriple=riscv64 -target-abi=lp64d -mattr=+experimental-v,+experimental-zfh,+f,+d -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=8 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX8
+; RUN: llc -mtriple=riscv32 -target-abi=ilp32d -mattr=+experimental-v,+experimental-zfh,+f,+d -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=1 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX1
+; RUN: llc -mtriple=riscv64 -target-abi=lp64d -mattr=+experimental-v,+experimental-zfh,+f,+d -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=1 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX1
+
+define void @gather_const_v8f16(<8 x half>* %x) {
+; CHECK-LABEL: gather_const_v8f16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 8
+; CHECK-NEXT:    vsetvli a1, a1, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 5
+; CHECK-NEXT:    vse16.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <8 x half>, <8 x half>* %x
+  %b = extractelement <8 x half> %a, i32 5
+  %c = insertelement <8 x half> undef, half %b, i32 0
+  %d = shufflevector <8 x half> %c, <8 x half> undef, <8 x i32> zeroinitializer
+  store <8 x half> %d, <8 x half>* %x
+  ret void
+}
+
+define void @gather_const_v4f32(<4 x float>* %x) {
+; CHECK-LABEL: gather_const_v4f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 4
+; CHECK-NEXT:    vsetvli a1, a1, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 2
+; CHECK-NEXT:    vse32.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <4 x float>, <4 x float>* %x
+  %b = extractelement <4 x float> %a, i32 2
+  %c = insertelement <4 x float> undef, float %b, i32 0
+  %d = shufflevector <4 x float> %c, <4 x float> undef, <4 x i32> zeroinitializer
+  store <4 x float> %d, <4 x float>* %x
+  ret void
+}
+
+define void @gather_const_v2f64(<2 x double>* %x) {
+; CHECK-LABEL: gather_const_v2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 2
+; CHECK-NEXT:    vsetvli a1, a1, e64,m1,ta,mu
+; CHECK-NEXT:    vle64.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 0
+; CHECK-NEXT:    vse64.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <2 x double>, <2 x double>* %x
+  %b = extractelement <2 x double> %a, i32 0
+  %c = insertelement <2 x double> undef, double %b, i32 0
+  %d = shufflevector <2 x double> %c, <2 x double> undef, <2 x i32> zeroinitializer
+  store <2 x double> %d, <2 x double>* %x
+  ret void
+}
+
+define void @gather_const_v64f16(<64 x half>* %x) {
+; LMULMAX8-LABEL: gather_const_v64f16:
+; LMULMAX8:       # %bb.0:
+; LMULMAX8-NEXT:    addi a1, zero, 64
+; LMULMAX8-NEXT:    vsetvli a1, a1, e16,m8,ta,mu
+; LMULMAX8-NEXT:    vle16.v v8, (a0)
+; LMULMAX8-NEXT:    addi a1, zero, 47
+; LMULMAX8-NEXT:    vrgather.vx v16, v8, a1
+; LMULMAX8-NEXT:    vse16.v v16, (a0)
+; LMULMAX8-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v64f16:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 80
+; LMULMAX1-NEXT:    addi a2, zero, 8
+; LMULMAX1-NEXT:    vsetvli a2, a2, e16,m1,ta,mu
+; LMULMAX1-NEXT:    vle16.v v25, (a1)
+; LMULMAX1-NEXT:    addi a6, a0, 16
+; LMULMAX1-NEXT:    addi a7, a0, 48
+; LMULMAX1-NEXT:    addi a4, a0, 32
+; LMULMAX1-NEXT:    addi a5, a0, 64
+; LMULMAX1-NEXT:    addi a2, a0, 112
+; LMULMAX1-NEXT:    addi a3, a0, 96
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 7
+; LMULMAX1-NEXT:    vse16.v v26, (a3)
+; LMULMAX1-NEXT:    vse16.v v26, (a2)
+; LMULMAX1-NEXT:    vse16.v v26, (a5)
+; LMULMAX1-NEXT:    vse16.v v26, (a1)
+; LMULMAX1-NEXT:    vse16.v v26, (a4)
+; LMULMAX1-NEXT:    vse16.v v26, (a7)
+; LMULMAX1-NEXT:    vse16.v v26, (a0)
+; LMULMAX1-NEXT:    vse16.v v26, (a6)
+; LMULMAX1-NEXT:    ret
+  %a = load <64 x half>, <64 x half>* %x
+  %b = extractelement <64 x half> %a, i32 47
+  %c = insertelement <64 x half> undef, half %b, i32 0
+  %d = shufflevector <64 x half> %c, <64 x half> undef, <64 x i32> zeroinitializer
+  store <64 x half> %d, <64 x half>* %x
+  ret void
+}
+
+define void @gather_const_v32f32(<32 x float>* %x) {
+; LMULMAX8-LABEL: gather_const_v32f32:
+; LMULMAX8:       # %bb.0:
+; LMULMAX8-NEXT:    addi a1, zero, 32
+; LMULMAX8-NEXT:    vsetvli a1, a1, e32,m8,ta,mu
+; LMULMAX8-NEXT:    vle32.v v8, (a0)
+; LMULMAX8-NEXT:    vrgather.vi v16, v8, 17
+; LMULMAX8-NEXT:    vse32.v v16, (a0)
+; LMULMAX8-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v32f32:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 64
+; LMULMAX1-NEXT:    addi a2, zero, 4
+; LMULMAX1-NEXT:    vsetvli a2, a2, e32,m1,ta,mu
+; LMULMAX1-NEXT:    vle32.v v25, (a1)
+; LMULMAX1-NEXT:    addi a6, a0, 16
+; LMULMAX1-NEXT:    addi a7, a0, 48
+; LMULMAX1-NEXT:    addi a4, a0, 32
+; LMULMAX1-NEXT:    addi a5, a0, 80
+; LMULMAX1-NEXT:    addi a2, a0, 112
+; LMULMAX1-NEXT:    addi a3, a0, 96
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 1
+; LMULMAX1-NEXT:    vse32.v v26, (a3)
+; LMULMAX1-NEXT:    vse32.v v26, (a2)
+; LMULMAX1-NEXT:    vse32.v v26, (a1)
+; LMULMAX1-NEXT:    vse32.v v26, (a5)
+; LMULMAX1-NEXT:    vse32.v v26, (a4)
+; LMULMAX1-NEXT:    vse32.v v26, (a7)
+; LMULMAX1-NEXT:    vse32.v v26, (a0)
+; LMULMAX1-NEXT:    vse32.v v26, (a6)
+; LMULMAX1-NEXT:    ret
+  %a = load <32 x float>, <32 x float>* %x
+  %b = extractelement <32 x float> %a, i32 17
+  %c = insertelement <32 x float> undef, float %b, i32 0
+  %d = shufflevector <32 x float> %c, <32 x float> undef, <32 x i32> zeroinitializer
+  store <32 x float> %d, <32 x float>* %x
+  ret void
+}
+
+define void @gather_const_v16f64(<16 x double>* %x) {
+; LMULMAX8-LABEL: gather_const_v16f64:
+; LMULMAX8:       # %bb.0:
+; LMULMAX8-NEXT:    addi a1, zero, 16
+; LMULMAX8-NEXT:    vsetvli a1, a1, e64,m8,ta,mu
+; LMULMAX8-NEXT:    vle64.v v8, (a0)
+; LMULMAX8-NEXT:    vrgather.vi v16, v8, 10
+; LMULMAX8-NEXT:    vse64.v v16, (a0)
+; LMULMAX8-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v16f64:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 80
+; LMULMAX1-NEXT:    addi a2, zero, 2
+; LMULMAX1-NEXT:    vsetvli a2, a2, e64,m1,ta,mu
+; LMULMAX1-NEXT:    vle64.v v25, (a1)
+; LMULMAX1-NEXT:    addi a6, a0, 16
+; LMULMAX1-NEXT:    addi a7, a0, 48
+; LMULMAX1-NEXT:    addi a4, a0, 32
+; LMULMAX1-NEXT:    addi a5, a0, 64
+; LMULMAX1-NEXT:    addi a2, a0, 112
+; LMULMAX1-NEXT:    addi a3, a0, 96
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 0
+; LMULMAX1-NEXT:    vse64.v v26, (a3)
+; LMULMAX1-NEXT:    vse64.v v26, (a2)
+; LMULMAX1-NEXT:    vse64.v v26, (a5)
+; LMULMAX1-NEXT:    vse64.v v26, (a1)
+; LMULMAX1-NEXT:    vse64.v v26, (a4)
+; LMULMAX1-NEXT:    vse64.v v26, (a7)
+; LMULMAX1-NEXT:    vse64.v v26, (a0)
+; LMULMAX1-NEXT:    vse64.v v26, (a6)
+; LMULMAX1-NEXT:    ret
+  %a = load <16 x double>, <16 x double>* %x
+  %b = extractelement <16 x double> %a, i32 10
+  %c = insertelement <16 x double> undef, double %b, i32 0
+  %d = shufflevector <16 x double> %c, <16 x double> undef, <16 x i32> zeroinitializer
+  store <16 x double> %d, <16 x double>* %x
+  ret void
+}

llvm/test/CodeGen/RISCV/rvv/fixed-vectors-int-vrgather.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=4 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX4
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=4 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX4
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=1 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX1
+; RUN: llc -mtriple=riscv64 -mattr=+experimental-v -verify-machineinstrs -riscv-v-vector-bits-min=128 -riscv-v-fixed-length-vector-lmul-max=1 -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK,LMULMAX1
+
+define void @gather_const_v16i8(<16 x i8>* %x) {
+; CHECK-LABEL: gather_const_v16i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 16
+; CHECK-NEXT:    vsetvli a1, a1, e8,m1,ta,mu
+; CHECK-NEXT:    vle8.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 12
+; CHECK-NEXT:    vse8.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <16 x i8>, <16 x i8>* %x
+  %b = extractelement <16 x i8> %a, i32 12
+  %c = insertelement <16 x i8> undef, i8 %b, i32 0
+  %d = shufflevector <16 x i8> %c, <16 x i8> undef, <16 x i32> zeroinitializer
+  store <16 x i8> %d, <16 x i8>* %x
+  ret void
+}
+
+define void @gather_const_v8i16(<8 x i16>* %x) {
+; CHECK-LABEL: gather_const_v8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 8
+; CHECK-NEXT:    vsetvli a1, a1, e16,m1,ta,mu
+; CHECK-NEXT:    vle16.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 5
+; CHECK-NEXT:    vse16.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <8 x i16>, <8 x i16>* %x
+  %b = extractelement <8 x i16> %a, i32 5
+  %c = insertelement <8 x i16> undef, i16 %b, i32 0
+  %d = shufflevector <8 x i16> %c, <8 x i16> undef, <8 x i32> zeroinitializer
+  store <8 x i16> %d, <8 x i16>* %x
+  ret void
+}
+
+define void @gather_const_v4i32(<4 x i32>* %x) {
+; CHECK-LABEL: gather_const_v4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 4
+; CHECK-NEXT:    vsetvli a1, a1, e32,m1,ta,mu
+; CHECK-NEXT:    vle32.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 3
+; CHECK-NEXT:    vse32.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <4 x i32>, <4 x i32>* %x
+  %b = extractelement <4 x i32> %a, i32 3
+  %c = insertelement <4 x i32> undef, i32 %b, i32 0
+  %d = shufflevector <4 x i32> %c, <4 x i32> undef, <4 x i32> zeroinitializer
+  store <4 x i32> %d, <4 x i32>* %x
+  ret void
+}
+
+define void @gather_const_v2i64(<2 x i64>* %x) {
+; CHECK-LABEL: gather_const_v2i64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    addi a1, zero, 2
+; CHECK-NEXT:    vsetvli a1, a1, e64,m1,ta,mu
+; CHECK-NEXT:    vle64.v v25, (a0)
+; CHECK-NEXT:    vrgather.vi v26, v25, 1
+; CHECK-NEXT:    vse64.v v26, (a0)
+; CHECK-NEXT:    ret
+  %a = load <2 x i64>, <2 x i64>* %x
+  %b = extractelement <2 x i64> %a, i32 1
+  %c = insertelement <2 x i64> undef, i64 %b, i32 0
+  %d = shufflevector <2 x i64> %c, <2 x i64> undef, <2 x i32> zeroinitializer
+  store <2 x i64> %d, <2 x i64>* %x
+  ret void
+}
+
+define void @gather_const_v64i8(<64 x i8>* %x) {
+; LMULMAX4-LABEL: gather_const_v64i8:
+; LMULMAX4:       # %bb.0:
+; LMULMAX4-NEXT:    addi a1, zero, 64
+; LMULMAX4-NEXT:    vsetvli a1, a1, e8,m4,ta,mu
+; LMULMAX4-NEXT:    vle8.v v28, (a0)
+; LMULMAX4-NEXT:    addi a1, zero, 32
+; LMULMAX4-NEXT:    vrgather.vx v8, v28, a1
+; LMULMAX4-NEXT:    vse8.v v8, (a0)
+; LMULMAX4-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v64i8:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 32
+; LMULMAX1-NEXT:    addi a2, zero, 16
+; LMULMAX1-NEXT:    vsetvli a2, a2, e8,m1,ta,mu
+; LMULMAX1-NEXT:    vle8.v v25, (a1)
+; LMULMAX1-NEXT:    addi a2, a0, 16
+; LMULMAX1-NEXT:    addi a3, a0, 48
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 0
+; LMULMAX1-NEXT:    vse8.v v26, (a1)
+; LMULMAX1-NEXT:    vse8.v v26, (a3)
+; LMULMAX1-NEXT:    vse8.v v26, (a0)
+; LMULMAX1-NEXT:    vse8.v v26, (a2)
+; LMULMAX1-NEXT:    ret
+  %a = load <64 x i8>, <64 x i8>* %x
+  %b = extractelement <64 x i8> %a, i32 32
+  %c = insertelement <64 x i8> undef, i8 %b, i32 0
+  %d = shufflevector <64 x i8> %c, <64 x i8> undef, <64 x i32> zeroinitializer
+  store <64 x i8> %d, <64 x i8>* %x
+  ret void
+}
+
+define void @gather_const_v16i16(<32 x i16>* %x) {
+; LMULMAX4-LABEL: gather_const_v16i16:
+; LMULMAX4:       # %bb.0:
+; LMULMAX4-NEXT:    addi a1, zero, 32
+; LMULMAX4-NEXT:    vsetvli a1, a1, e16,m4,ta,mu
+; LMULMAX4-NEXT:    vle16.v v28, (a0)
+; LMULMAX4-NEXT:    vrgather.vi v8, v28, 25
+; LMULMAX4-NEXT:    vse16.v v8, (a0)
+; LMULMAX4-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v16i16:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 48
+; LMULMAX1-NEXT:    addi a2, zero, 8
+; LMULMAX1-NEXT:    vsetvli a2, a2, e16,m1,ta,mu
+; LMULMAX1-NEXT:    vle16.v v25, (a1)
+; LMULMAX1-NEXT:    addi a2, a0, 16
+; LMULMAX1-NEXT:    addi a3, a0, 32
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 1
+; LMULMAX1-NEXT:    vse16.v v26, (a3)
+; LMULMAX1-NEXT:    vse16.v v26, (a1)
+; LMULMAX1-NEXT:    vse16.v v26, (a0)
+; LMULMAX1-NEXT:    vse16.v v26, (a2)
+; LMULMAX1-NEXT:    ret
+  %a = load <32 x i16>, <32 x i16>* %x
+  %b = extractelement <32 x i16> %a, i32 25
+  %c = insertelement <32 x i16> undef, i16 %b, i32 0
+  %d = shufflevector <32 x i16> %c, <32 x i16> undef, <32 x i32> zeroinitializer
+  store <32 x i16> %d, <32 x i16>* %x
+  ret void
+}
+
+define void @gather_const_v16i32(<16 x i32>* %x) {
+; LMULMAX4-LABEL: gather_const_v16i32:
+; LMULMAX4:       # %bb.0:
+; LMULMAX4-NEXT:    addi a1, zero, 16
+; LMULMAX4-NEXT:    vsetvli a1, a1, e32,m4,ta,mu
+; LMULMAX4-NEXT:    vle32.v v28, (a0)
+; LMULMAX4-NEXT:    vrgather.vi v8, v28, 9
+; LMULMAX4-NEXT:    vse32.v v8, (a0)
+; LMULMAX4-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v16i32:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 32
+; LMULMAX1-NEXT:    addi a2, zero, 4
+; LMULMAX1-NEXT:    vsetvli a2, a2, e32,m1,ta,mu
+; LMULMAX1-NEXT:    vle32.v v25, (a1)
+; LMULMAX1-NEXT:    addi a2, a0, 16
+; LMULMAX1-NEXT:    addi a3, a0, 48
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 1
+; LMULMAX1-NEXT:    vse32.v v26, (a1)
+; LMULMAX1-NEXT:    vse32.v v26, (a3)
+; LMULMAX1-NEXT:    vse32.v v26, (a0)
+; LMULMAX1-NEXT:    vse32.v v26, (a2)
+; LMULMAX1-NEXT:    ret
+  %a = load <16 x i32>, <16 x i32>* %x
+  %b = extractelement <16 x i32> %a, i32 9
+  %c = insertelement <16 x i32> undef, i32 %b, i32 0
+  %d = shufflevector <16 x i32> %c, <16 x i32> undef, <16 x i32> zeroinitializer
+  store <16 x i32> %d, <16 x i32>* %x
+  ret void
+}
+
+define void @gather_const_v8i64(<8 x i64>* %x) {
+; LMULMAX4-LABEL: gather_const_v8i64:
+; LMULMAX4:       # %bb.0:
+; LMULMAX4-NEXT:    addi a1, zero, 8
+; LMULMAX4-NEXT:    vsetvli a1, a1, e64,m4,ta,mu
+; LMULMAX4-NEXT:    vle64.v v28, (a0)
+; LMULMAX4-NEXT:    vrgather.vi v8, v28, 3
+; LMULMAX4-NEXT:    vse64.v v8, (a0)
+; LMULMAX4-NEXT:    ret
+;
+; LMULMAX1-LABEL: gather_const_v8i64:
+; LMULMAX1:       # %bb.0:
+; LMULMAX1-NEXT:    addi a1, a0, 16
+; LMULMAX1-NEXT:    addi a2, zero, 2
+; LMULMAX1-NEXT:    vsetvli a2, a2, e64,m1,ta,mu
+; LMULMAX1-NEXT:    vle64.v v25, (a1)
+; LMULMAX1-NEXT:    addi a2, a0, 48
+; LMULMAX1-NEXT:    addi a3, a0, 32
+; LMULMAX1-NEXT:    vrgather.vi v26, v25, 1
+; LMULMAX1-NEXT:    vse64.v v26, (a3)
+; LMULMAX1-NEXT:    vse64.v v26, (a2)
+; LMULMAX1-NEXT:    vse64.v v26, (a0)
+; LMULMAX1-NEXT:    vse64.v v26, (a1)
+; LMULMAX1-NEXT:    ret
+  %a = load <8 x i64>, <8 x i64>* %x
+  %b = extractelement <8 x i64> %a, i32 3
+  %c = insertelement <8 x i64> undef, i64 %b, i32 0
+  %d = shufflevector <8 x i64> %c, <8 x i64> undef, <8 x i32> zeroinitializer
+  store <8 x i64> %d, <8 x i64>* %x
+  ret void
+}