[mips][msa] Mask vectors holding shift amounts

[smaksimovic]

Summary

Masked vectors which hold shift amounts when creating following nodes: ISD::SHL, ISD::SRL or ISD::SRA.
Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset.

For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements
will be interpreted as the shift amount/bit position modulo the size of the element in bits.

The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated,
but are instead optimized away.
In this case, having shift amounts that are either negative or greater than the element bit size results in
generation of incorrect results when constant folding.

We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the
final result is correct before placed into the constant pool.

[efriedma]

This is obviously not the correct fix: you need to mask the shift amount whether or not you can prove the shift amount is constant in lowerINTRINSIC_WO_CHAIN.

[smaksimovic]

Hello, thanks for the comment.

Made a change introducing an ISD::AND node which is now used to truncate the vector elements which hold shift amounts regardless of the vector having constants or not.

One downside we're experiencing now is that although we produce the code that executes correctly, and.v instruction is generated in addition to sll, where and.v is superfluous since sll will handle both negative and those shift amounts that are larger by value than the vector element size correctly.

Ideally, we'd like the sll and and.v to be optimized away in case all of the operands are constant values.
Additionally, this could be restricted to -O2, since on lower optimization levels shift instructions are generated anyway, hence and.v will be redundant in that situation as well.

Any tips on handling these two issues?

[efriedma]

You basically need to pattern-match the ISD::AND+ISD::SHL to the appropriate native instruction. Given that the AND mask is going to get lowered to a constant-pool load (I think?), it's tricky to match with TableGen patterns; probably easiest to add a target-specific MIPSISD::VSHL etc. which masks the shift amount. The x86 backend does something similar.

[sdardis]

This latest diff is too general. For the constant case we can perform the bit masking at compile time to reformulate the constant into something that LLVM's IR will treat as legal and that matches what the hardware will do.

For the general case we can elide the '(and operand, (build_vector <ElementSizeInBits -1>...))' with the following:

def immi32Cst31 : ImmLeaf<i32, [{return isUInt<32>(Imm) && Imm == 31;}]>;

def : MSAPat<
  (v4i32 (shl v4i32:$ws, (v4i32 (and v4i32:$wt,
                                     (build_vector immi32Cst31, immi32Cst31,
                                                   immi32Cst31, immi32Cst31))))),
  (v4i32 (SLL_W v4i32:$ws, v4i32:$wt))>;

(That example may need some reformatting.)

That example which I've briefly tested removes the and SDNode during ISel. You can modify that pattern into a multiclass and then instantiate it for the relevant instructions.

In principle we should cover the .b and .h cases as well.

lib/Target/Mips/MipsSEISelLowering.cpp

1556–1560 ↗

(On Diff #93624)

This hunk only covers the non-constant case. When the operand to a logical ISD vector node--which implicitly masks the lower bits--is a ConstantSDnode, we should instead reformulate the constant so that it only contains bits that the MSA instructions will look at.

[efriedma]

lib/Target/Mips/MipsSEISelLowering.cpp

1556–1560 ↗

(On Diff #93624)

Doesn't constant folding cover the constant case?

[sdardis]

@efriedma was right about constant folding taking care of the constant case, so you can ignore my comments about that.

Some more comments inlined, mostly on the tests.

lib/Target/Mips/MipsSEISelLowering.cpp

1556–1560 ↗

(On Diff #93624)

You right, it does. I assumed it wouldn't.

test/CodeGen/Mips/msa/3r-s.ll

134 ↗

(On Diff #93624)

For the test cases you've changed, can you update the CHECK: <function name>: to CHECK-LABEL: ?

143 ↗

(On Diff #93624)

These extra ; can be removed in the changed cases as well.

643 ↗

(On Diff #93624)

This can be removed.

test/CodeGen/Mips/msa/shift_constant_pool.ll

4–5 ↗

(On Diff #93624)

Rather than using CHECK-ALL, you can just use ALL for brevity. The multiple -check-prefix s can be combined into --check-prefixes=ALL,MIPS64 / --check-prefixes=ALL,MIPS32

24 ↗

(On Diff #93624)

This should ALL-LABEL: llvm_mips_sll_w_test_const_vec:

33–34 ↗

(On Diff #93624)

These two lines are unnecessary.

52 ↗

(On Diff #93624)

See my comments on llvm_mips_sll_w_test_const_vec about ALL-LABEL and the last two lines which can be dropped.

80 ↗

(On Diff #93624)

See my comments on llvm_mips_sll_w_test_const_vec about ALL-LABEL and the last two lines which can be dropped.

[smaksimovic]

• Added classes covering patterns which remove and nodes for sll, srl and sra instructions in MipsMSAInstrInfo.td
• Modified the existing patterns for bclr, bneg, bset instructions to match the additional and node which we now create
• Dropped changes in test/CodeGen/Mips/msa/3r-s.ll, since we do not generate and instructions now, so no changes required
• Added a test which specifically checks for said and instructions not being present in the resulting code

[sdardis]

Testing this shows that test/CodeGen/Mips/msa/bitwise.ll is now failing as the bitwise instructions are no longer selected.

lib/Target/Mips/MipsMSAInstrInfo.td

355 ↗

(On Diff #94800)

This predicate is unused.

371–410 ↗

(On Diff #94800)

These changes are regressing the general bitwise ISel logic for these instructions, as now their patterns are matching what the intrinsics produce when those intrinsics are lowered.

You'll probably need to duplicate the patterns instead of changing them.

[smaksimovic]

Addressed review comments.

[sdardis]

LGTM, you'll need @efriedma to accept this as well before this revision can be closed.

lib/Target/Mips/MipsMSAInstrInfo.td

3843–3861 ↗

(On Diff #95552)

Nit: These can be redone as a multiclass pattern parametrized over the SDNode and instruction, similar to what you did with the MSAShiftPats.

[smaksimovic]

Redefined bset and bneg patterns via multiclass as suggested.

[efriedma]

LGTM.

llvm/trunk/lib/Target/Mips/MipsMSAInstrInfo.td

def : MipsPat<(MipsTruncIntFP MSA128F16:$ws),
                (TRUNC_W_D64 (MSA_FP_EXTEND_D_PSEUDO MSA128F16:$ws))>;
 
  def : MipsPat<(MipsFPCmp MSA128F16:$ws, MSA128F16:$wt, imm:$cond),
                (FCMP_S32 (MSA_FP_EXTEND_W_PSEUDO MSA128F16:$ws),
                          (MSA_FP_EXTEND_W_PSEUDO MSA128F16:$wt), imm:$cond)>,
        ISA_MIPS1_NOT_32R6_64R6;
 }
 
+def vsplati64_imm_eq_63 : PatLeaf<(bitconvert (v4i32 (build_vector))), [{
+  APInt Imm;
+  SDNode *BV = N->getOperand(0).getNode();
+  EVT EltTy = N->getValueType(0).getVectorElementType();
+
+  return selectVSplat(BV, Imm, EltTy.getSizeInBits()) &&
+         Imm.getBitWidth() == EltTy.getSizeInBits() && Imm == 63;
+}]>;
+
+def immi32Cst7  : ImmLeaf<i32, [{return isUInt<32>(Imm) && Imm == 7;}]>;
+def immi32Cst15 : ImmLeaf<i32, [{return isUInt<32>(Imm) && Imm == 15;}]>;
+def immi32Cst31 : ImmLeaf<i32, [{return isUInt<32>(Imm) && Imm == 31;}]>;
+
+def vsplati8imm7 :   PatFrag<(ops node:$wt),
+                             (and node:$wt, (vsplati8 immi32Cst7))>;
+def vsplati16imm15 : PatFrag<(ops node:$wt),
+                             (and node:$wt, (vsplati16 immi32Cst15))>;
+def vsplati32imm31 : PatFrag<(ops node:$wt),
+                             (and node:$wt, (vsplati32 immi32Cst31))>;
+def vsplati64imm63 : PatFrag<(ops node:$wt),
+                             (and node:$wt, vsplati64_imm_eq_63)>;
+
+class MSAShiftPat<SDNode Node, ValueType VT, MSAInst Insn, dag Vec> :
+  MSAPat<(VT (Node VT:$ws, (VT (and VT:$wt, Vec)))),
+         (VT (Insn VT:$ws, VT:$wt))>;
+
+class MSABitPat<SDNode Node, ValueType VT, MSAInst Insn, PatFrag Frag> :
+  MSAPat<(VT (Node VT:$ws, (shl vsplat_imm_eq_1, (Frag VT:$wt)))),
+         (VT (Insn VT:$ws, VT:$wt))>;
+
+multiclass MSAShiftPats<SDNode Node, string Insn> {
+  def : MSAShiftPat<Node, v16i8, !cast<MSAInst>(Insn#_B),
+                    (vsplati8 immi32Cst7)>;
+  def : MSAShiftPat<Node, v8i16, !cast<MSAInst>(Insn#_H),
+                    (vsplati16 immi32Cst15)>;
+  def : MSAShiftPat<Node, v4i32, !cast<MSAInst>(Insn#_W),
+                    (vsplati32 immi32Cst31)>;
+  def : MSAPat<(v2i64 (Node v2i64:$ws, (v2i64 (and v2i64:$wt,
+                                                   vsplati64_imm_eq_63)))),
+               (v2i64 (!cast<MSAInst>(Insn#_D) v2i64:$ws, v2i64:$wt))>;
+}
+
+multiclass MSABitPats<SDNode Node, string Insn> {
+  def : MSABitPat<Node, v16i8, !cast<MSAInst>(Insn#_B), vsplati8imm7>;
+  def : MSABitPat<Node, v8i16, !cast<MSAInst>(Insn#_H), vsplati16imm15>;
+  def : MSABitPat<Node, v4i32, !cast<MSAInst>(Insn#_W), vsplati32imm31>;
+  def : MSAPat<(Node v2i64:$ws, (shl (v2i64 vsplati64_imm_eq_1),
+                                     (vsplati64imm63 v2i64:$wt))),
+               (v2i64 (!cast<MSAInst>(Insn#_D) v2i64:$ws, v2i64:$wt))>;
+}
+
+defm : MSAShiftPats<shl, "SLL">;
+defm : MSAShiftPats<srl, "SRL">;
+defm : MSAShiftPats<sra, "SRA">;
+defm : MSABitPats<xor, "BNEG">;
+defm : MSABitPats<or, "BSET">;
+
+def : MSAPat<(and v16i8:$ws, (xor (shl vsplat_imm_eq_1,
+                                       (vsplati8imm7 v16i8:$wt)),
+                                  immAllOnesV)),
+             (v16i8 (BCLR_B v16i8:$ws, v16i8:$wt))>;
+def : MSAPat<(and v8i16:$ws, (xor (shl vsplat_imm_eq_1,
+                                       (vsplati16imm15 v8i16:$wt)),
+                             immAllOnesV)),
+             (v8i16 (BCLR_H v8i16:$ws, v8i16:$wt))>;
+def : MSAPat<(and v4i32:$ws, (xor (shl vsplat_imm_eq_1,
+                                       (vsplati32imm31 v4i32:$wt)),
+                             immAllOnesV)),
+             (v4i32 (BCLR_W v4i32:$ws, v4i32:$wt))>;
+def : MSAPat<(and v2i64:$ws, (xor (shl (v2i64 vsplati64_imm_eq_1),
+                                       (vsplati64imm63 v2i64:$wt)),
+                                  (bitconvert (v4i32 immAllOnesV)))),
+             (v2i64 (BCLR_D v2i64:$ws, v2i64:$wt))>;
+
 // Vector extraction with fixed index.
 //
 // Extracting 32-bit values on MSA32 should always use COPY_S_W rather than
 // COPY_U_W, even for the zero-extended case. This is because our forward
 // compatibility strategy is to consider registers to be infinitely
 // sign-extended so that a MIPS64 can execute MIPS32 code without getting
 // different register values.
 def : MSAPat<(vextract_zext_i32 (v4i32 MSA128W:$ws), immZExt2Ptr:$idx),

llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp

if (!Exp2Imm.getNode()) {
 
     Exp2Imm = DAG.getNode(ISD::SHL, DL, VecTy, DAG.getConstant(1, DL, VecTy),
                           Exp2Imm);
   }
 
   return DAG.getNode(Opc, DL, VecTy, Op->getOperand(1), Exp2Imm);
 }
 
+static SDValue truncateVecElts(SDValue Op, SelectionDAG &DAG) {
+  SDLoc DL(Op);
+  EVT ResTy = Op->getValueType(0);
+  SDValue Vec = Op->getOperand(2);
+  bool BigEndian = !DAG.getSubtarget().getTargetTriple().isLittleEndian();
+  MVT ResEltTy = ResTy == MVT::v2i64 ? MVT::i64 : MVT::i32;
+  SDValue ConstValue = DAG.getConstant(Vec.getScalarValueSizeInBits() - 1,
+                                       DL, ResEltTy);
+  SDValue SplatVec = getBuildVectorSplat(ResTy, ConstValue, BigEndian, DAG);
+
+  return DAG.getNode(ISD::AND, DL, ResTy, Vec, SplatVec);
+}
+
 static SDValue lowerMSABitClear(SDValue Op, SelectionDAG &DAG) {
   EVT ResTy = Op->getValueType(0);
   SDLoc DL(Op);
   SDValue One = DAG.getConstant(1, DL, ResTy);
-  SDValue Bit = DAG.getNode(ISD::SHL, DL, ResTy, One, Op->getOperand(2));
+  SDValue Bit = DAG.getNode(ISD::SHL, DL, ResTy, One, truncateVecElts(Op, DAG));
 
   return DAG.getNode(ISD::AND, DL, ResTy, Op->getOperand(1),
                      DAG.getNOT(DL, Bit, ResTy));
 }
 
 static SDValue lowerMSABitClearImm(SDValue Op, SelectionDAG &DAG) {
   SDLoc DL(Op);
   EVT ResTy = Op->getValueType(0);
SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::mips_bneg_h:
   case Intrinsic::mips_bneg_w:
   case Intrinsic::mips_bneg_d: {
     EVT VecTy = Op->getValueType(0);
     SDValue One = DAG.getConstant(1, DL, VecTy);
 
     return DAG.getNode(ISD::XOR, DL, VecTy, Op->getOperand(1),
                        DAG.getNode(ISD::SHL, DL, VecTy, One,
-                                   Op->getOperand(2)));
+                                   truncateVecElts(Op, DAG)));
   }
   case Intrinsic::mips_bnegi_b:
   case Intrinsic::mips_bnegi_h:
   case Intrinsic::mips_bnegi_w:
   case Intrinsic::mips_bnegi_d:
     return lowerMSABinaryBitImmIntr(Op, DAG, ISD::XOR, Op->getOperand(2),
                                     !Subtarget.isLittle());
   case Intrinsic::mips_bnz_b:
SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::mips_bset_h:
   case Intrinsic::mips_bset_w:
   case Intrinsic::mips_bset_d: {
     EVT VecTy = Op->getValueType(0);
     SDValue One = DAG.getConstant(1, DL, VecTy);
 
     return DAG.getNode(ISD::OR, DL, VecTy, Op->getOperand(1),
                        DAG.getNode(ISD::SHL, DL, VecTy, One,
-                                   Op->getOperand(2)));
+                                   truncateVecElts(Op, DAG)));
   }
   case Intrinsic::mips_bseti_b:
   case Intrinsic::mips_bseti_h:
   case Intrinsic::mips_bseti_w:
   case Intrinsic::mips_bseti_d:
     return lowerMSABinaryBitImmIntr(Op, DAG, ISD::OR, Op->getOperand(2),
                                     !Subtarget.isLittle());
   case Intrinsic::mips_bz_b:
if (Value < 0 || Value > Max)
       report_fatal_error("Immediate out of range");
     return SDValue();
   }
   case Intrinsic::mips_sll_b:
   case Intrinsic::mips_sll_h:
   case Intrinsic::mips_sll_w:
   case Intrinsic::mips_sll_d:
     return DAG.getNode(ISD::SHL, DL, Op->getValueType(0), Op->getOperand(1),
-                       Op->getOperand(2));
+                       truncateVecElts(Op, DAG));
   case Intrinsic::mips_slli_b:
   case Intrinsic::mips_slli_h:
   case Intrinsic::mips_slli_w:
   case Intrinsic::mips_slli_d:
     return DAG.getNode(ISD::SHL, DL, Op->getValueType(0),
                        Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
   case Intrinsic::mips_splat_b:
   case Intrinsic::mips_splat_h:
case Intrinsic::mips_splati_d:
     return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
                        lowerMSASplatImm(Op, 2, DAG), Op->getOperand(1),
                        Op->getOperand(1));
   case Intrinsic::mips_sra_b:
   case Intrinsic::mips_sra_h:
   case Intrinsic::mips_sra_w:
   case Intrinsic::mips_sra_d:
     return DAG.getNode(ISD::SRA, DL, Op->getValueType(0), Op->getOperand(1),
-                       Op->getOperand(2));
+                       truncateVecElts(Op, DAG));
   case Intrinsic::mips_srai_b:
   case Intrinsic::mips_srai_h:
   case Intrinsic::mips_srai_w:
   case Intrinsic::mips_srai_d:
     return DAG.getNode(ISD::SRA, DL, Op->getValueType(0),
                        Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
   case Intrinsic::mips_srari_b:
   case Intrinsic::mips_srari_h:
if (Value < 0 || Value > Max)
       report_fatal_error("Immediate out of range");
     return SDValue();
   }
   case Intrinsic::mips_srl_b:
   case Intrinsic::mips_srl_h:
   case Intrinsic::mips_srl_w:
   case Intrinsic::mips_srl_d:
     return DAG.getNode(ISD::SRL, DL, Op->getValueType(0), Op->getOperand(1),
-                       Op->getOperand(2));
+                       truncateVecElts(Op, DAG));
   case Intrinsic::mips_srli_b:
   case Intrinsic::mips_srli_h:
   case Intrinsic::mips_srli_w:
   case Intrinsic::mips_srli_d:
     return DAG.getNode(ISD::SRL, DL, Op->getValueType(0),
                        Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
   case Intrinsic::mips_srlri_b:
   case Intrinsic::mips_srlri_h:

llvm/trunk/test/CodeGen/Mips/msa/shift_constant_pool.ll

+; Test whether the following functions, with vectors featuring negative or values larger than the element
+; bit size have their results of operations generated correctly when placed into constant pools
+
+; RUN: llc -march=mips64 -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck -check-prefixes=ALL,MIPS64 %s
+; RUN: llc -march=mips -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck -check-prefixes=ALL,MIPS32 %s
+; RUN: llc -march=mips64el -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck -check-prefixes=ALL,MIPS64 %s
+; RUN: llc -march=mipsel -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck -check-prefixes=ALL,MIPS32 %s
+
+@llvm_mips_bclr_w_test_const_vec_res = global <4 x i32> zeroinitializer, align 16
+
+define void @llvm_mips_bclr_w_test_const_vec() nounwind {
+entry:
+  %0 = tail call <4 x i32> @llvm.mips.bclr.w(<4 x i32> <i32 2147483649, i32 2147483649, i32 7, i32 7>, <4 x i32> <i32 -1, i32 31, i32 2, i32 34>)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_bclr_w_test_const_vec_res
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.bclr.w(<4 x i32>, <4 x i32>) nounwind
+
+; MIPS32: [[LABEL:\$CPI[0-9]+_[0-9]+]]:
+; MIPS64: [[LABEL:\.LCPI[0-9]+_[0-9]+]]:
+; ALL:  .4byte  1                       # 0x1
+; ALL:  .4byte  1                       # 0x1
+; ALL:  .4byte  3                       # 0x3
+; ALL:  .4byte  3                       # 0x3
+; ALL-LABEL: llvm_mips_bclr_w_test_const_vec:
+; MIPS32: lw $[[R2:[0-9]+]], %got([[LABEL]])($[[R1:[0-9]+]])
+; MIPS32: addiu $[[R2]], $[[R2]], %lo([[LABEL]])
+; MIPS32: lw $[[R3:[0-9]+]], %got(llvm_mips_bclr_w_test_const_vec_res)($[[R1]])
+; MIPS64: ld $[[R2:[0-9]+]], %got_page([[LABEL]])($[[R1:[0-9]+]])
+; MIPS64: daddiu $[[R2]], $[[R2]], %got_ofst([[LABEL]])
+; MIPS64: ld $[[R3:[0-9]+]], %got_disp(llvm_mips_bclr_w_test_const_vec_res)($[[R1]])
+; ALL: ld.w $w0, 0($[[R2]])
+; ALL: st.w $w0, 0($[[R3]])
+
+
+@llvm_mips_bneg_w_test_const_vec_res = global <4 x i32> zeroinitializer, align 16
+
+define void @llvm_mips_bneg_w_test_const_vec() nounwind {
+entry:
+  %0 = tail call <4 x i32> @llvm.mips.bneg.w(<4 x i32> <i32 2147483649, i32 2147483649, i32 7, i32 7>, <4 x i32> <i32 -1, i32 31, i32 2, i32 34>)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_bneg_w_test_const_vec_res
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.bneg.w(<4 x i32>, <4 x i32>) nounwind
+
+; MIPS32: [[LABEL:\$CPI[0-9]+_[0-9]+]]:
+; MIPS64: [[LABEL:\.LCPI[0-9]+_[0-9]+]]:
+; ALL:  .4byte  1                       # 0x1
+; ALL:  .4byte  1                       # 0x1
+; ALL:  .4byte  3                       # 0x3
+; ALL:  .4byte  3                       # 0x3
+; ALL-LABEL: llvm_mips_bneg_w_test_const_vec:
+; MIPS32: lw $[[R2:[0-9]+]], %got([[LABEL]])($[[R1:[0-9]+]])
+; MIPS32: addiu $[[R2]], $[[R2]], %lo([[LABEL]])
+; MIPS32: lw $[[R3:[0-9]+]], %got(llvm_mips_bneg_w_test_const_vec_res)($[[R1]])
+; MIPS64: ld $[[R2:[0-9]+]], %got_page([[LABEL]])($[[R1:[0-9]+]])
+; MIPS64: daddiu $[[R2]], $[[R2]], %got_ofst([[LABEL]])
+; MIPS64: ld $[[R3:[0-9]+]], %got_disp(llvm_mips_bneg_w_test_const_vec_res)($[[R1]])
+; ALL: ld.w $w0, 0($[[R2]])
+; ALL: st.w $w0, 0($[[R3]])
+
+
+@llvm_mips_bset_w_test_const_vec_res = global <4 x i32> zeroinitializer, align 16
+
+define void @llvm_mips_bset_w_test_const_vec() nounwind {
+entry:
+  %0 = tail call <4 x i32> @llvm.mips.bset.w(<4 x i32> <i32 0, i32 0, i32 0, i32 0>, <4 x i32> <i32 -1, i32 31, i32 2, i32 34>)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_bset_w_test_const_vec_res
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.bset.w(<4 x i32>, <4 x i32>) nounwind
+
+; MIPS32: [[LABEL:\$CPI[0-9]+_[0-9]+]]:
+; MIPS64: [[LABEL:\.LCPI[0-9]+_[0-9]+]]:
+; ALL:  .4byte  2147483648              # 0x80000000
+; ALL:  .4byte  2147483648              # 0x80000000
+; ALL:  .4byte  4                       # 0x4
+; ALL:  .4byte  4                       # 0x4
+; ALL-LABEL: llvm_mips_bset_w_test_const_vec:
+; MIPS32: lw $[[R2:[0-9]+]], %got([[LABEL]])($[[R1:[0-9]+]])
+; MIPS32: addiu $[[R2]], $[[R2]], %lo([[LABEL]])
+; MIPS32: lw $[[R3:[0-9]+]], %got(llvm_mips_bset_w_test_const_vec_res)($[[R1]])
+; MIPS64: ld $[[R2:[0-9]+]], %got_page([[LABEL]])($[[R1:[0-9]+]])
+; MIPS64: daddiu $[[R2]], $[[R2]], %got_ofst([[LABEL]])
+; MIPS64: ld $[[R3:[0-9]+]], %got_disp(llvm_mips_bset_w_test_const_vec_res)($[[R1]])
+; ALL: ld.w $w0, 0($[[R2]])
+; ALL: st.w $w0, 0($[[R3]])
+
+@llvm_mips_sll_w_test_const_vec_res = global <4 x i32> zeroinitializer, align 16
+
+define void @llvm_mips_sll_w_test_const_vec() nounwind {
+entry:
+  %0 = tail call <4 x i32> @llvm.mips.sll.w(<4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 -1, i32 31, i32 2, i32 34>)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_sll_w_test_const_vec_res
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.sll.w(<4 x i32>, <4 x i32>) nounwind
+
+; MIPS32: [[LABEL:\$CPI[0-9]+_[0-9]+]]:
+; MIPS64: [[LABEL:\.LCPI[0-9]+_[0-9]+]]:
+; ALL: .4byte 2147483648              # 0x80000000
+; ALL: .4byte 2147483648              # 0x80000000
+; ALL: .4byte 4                       # 0x4
+; ALL: .4byte 4                       # 0x4
+; ALL-LABEL: llvm_mips_sll_w_test_const_vec:
+; MIPS32: lw $[[R2:[0-9]+]], %got([[LABEL]])($[[R1:[0-9]+]])
+; MIPS32: addiu $[[R2]], $[[R2]], %lo([[LABEL]])
+; MIPS32: lw $[[R3:[0-9]+]], %got(llvm_mips_sll_w_test_const_vec_res)($[[R1]])
+; MIPS64: ld $[[R2:[0-9]+]], %got_page([[LABEL]])($[[R1:[0-9]+]])
+; MIPS64: daddiu $[[R2]], $[[R2]], %got_ofst([[LABEL]])
+; MIPS64: ld $[[R3:[0-9]+]], %got_disp(llvm_mips_sll_w_test_const_vec_res)($[[R1]])
+; ALL: ld.w $w0, 0($[[R2]])
+; ALL: st.w $w0, 0($[[R3]])
+
+@llvm_mips_sra_w_test_const_vec_res = global <4 x i32> zeroinitializer, align 16
+
+define void @llvm_mips_sra_w_test_const_vec() nounwind {
+entry:
+  %0 = tail call <4 x i32> @llvm.mips.sra.w(<4 x i32> <i32 -16, i32 16, i32 16, i32 16>, <4 x i32> <i32 2, i32 -30, i32 33, i32 1>)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_sra_w_test_const_vec_res
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.sra.w(<4 x i32>, <4 x i32>) nounwind
+
+; MIPS32: [[LABEL:\$CPI[0-9]+_[0-9]+]]:
+; MIPS64: [[LABEL:\.LCPI[0-9]+_[0-9]+]]:
+; ALL: .4byte 4294967292              # 0xfffffffc
+; ALL: .4byte 4                       # 0x4
+; ALL: .4byte 8                       # 0x8
+; ALL: .4byte 8                       # 0x8
+; ALL-LABEL: llvm_mips_sra_w_test_const_vec:
+; MIPS32: lw $[[R2:[0-9]+]], %got([[LABEL]])($[[R1:[0-9]+]])
+; MIPS32: addiu $[[R2]], $[[R2]], %lo([[LABEL]])
+; MIPS32: lw $[[R3:[0-9]+]], %got(llvm_mips_sra_w_test_const_vec_res)($[[R1]])
+; MIPS64: ld $[[R2:[0-9]+]], %got_page([[LABEL]])($[[R1:[0-9]+]])
+; MIPS64: daddiu $[[R2]], $[[R2]], %got_ofst([[LABEL]])
+; MIPS64: ld $[[R3:[0-9]+]], %got_disp(llvm_mips_sra_w_test_const_vec_res)($[[R1]])
+; ALL: ld.w $w0, 0($[[R2]])
+; ALL: st.w $w0, 0($[[R3]])
+
+@llvm_mips_srl_w_test_const_vec_res = global <4 x i32> zeroinitializer, align 16
+
+define void @llvm_mips_srl_w_test_const_vec() nounwind {
+entry:
+  %0 = tail call <4 x i32> @llvm.mips.srl.w(<4 x i32> <i32 -16, i32 16, i32 16, i32 16>, <4 x i32> <i32 2, i32 -30, i32 33, i32 1>)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_srl_w_test_const_vec_res
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.srl.w(<4 x i32>, <4 x i32>) nounwind
+
+; MIPS32: [[LABEL:\$CPI[0-9]+_[0-9]+]]:
+; MIPS64: [[LABEL:\.LCPI[0-9]+_[0-9]+]]:
+; ALL: .4byte 1073741820              # 0x3ffffffc
+; ALL: .4byte 4                       # 0x4
+; ALL: .4byte 8                       # 0x8
+; ALL: .4byte 8                       # 0x8
+; ALL-LABEL: llvm_mips_srl_w_test_const_vec:
+; MIPS32: lw $[[R2:[0-9]+]], %got([[LABEL]])($[[R1:[0-9]+]])
+; MIPS32: addiu $[[R2]], $[[R2]], %lo([[LABEL]])
+; MIPS32: lw $[[R3:[0-9]+]], %got(llvm_mips_srl_w_test_const_vec_res)($[[R1]])
+; MIPS64: ld $[[R2:[0-9]+]], %got_page([[LABEL]])($[[R1:[0-9]+]])
+; MIPS64: daddiu $[[R2]], $[[R2]], %got_ofst([[LABEL]])
+; MIPS64: ld $[[R3:[0-9]+]], %got_disp(llvm_mips_srl_w_test_const_vec_res)($[[R1]])
+; ALL: ld.w $w0, 0($[[R2]])
+; ALL: st.w $w0, 0($[[R3]])

llvm/trunk/test/CodeGen/Mips/msa/shift_no_and.ll

+; Test the absence of the andi.b / and.v instructions
+
+; RUN: llc -march=mips -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck %s
+; RUN: llc -march=mipsel -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck %s
+
+@llvm_mips_bclr_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
+@llvm_mips_bclr_b_ARG2 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
+@llvm_mips_bclr_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_bclr_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_bclr_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_bclr_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.bclr.b(<16 x i8> %0, <16 x i8> %1)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_bclr_b_RES
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.bclr.b(<16 x i8>, <16 x i8>) nounwind
+
+; CHECK-LABEL: llvm_mips_bclr_b_test:
+; CHECK-NOT: andi.b
+; CHECK: bclr.b
+
+@llvm_mips_bclr_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
+@llvm_mips_bclr_h_ARG2 = global <8 x i16> <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, align 16
+@llvm_mips_bclr_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_bclr_h_test() nounwind {
+entry:
+  %0 = load <8 x i16>, <8 x i16>* @llvm_mips_bclr_h_ARG1
+  %1 = load <8 x i16>, <8 x i16>* @llvm_mips_bclr_h_ARG2
+  %2 = tail call <8 x i16> @llvm.mips.bclr.h(<8 x i16> %0, <8 x i16> %1)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_bclr_h_RES
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.bclr.h(<8 x i16>, <8 x i16>) nounwind
+
+; CHECK-LABEL: llvm_mips_bclr_h_test:
+; CHECK-NOT: and.v
+; CHECK: bclr.h
+
+@llvm_mips_bclr_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
+@llvm_mips_bclr_w_ARG2 = global <4 x i32> <i32 4, i32 5, i32 6, i32 7>, align 16
+@llvm_mips_bclr_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_bclr_w_test() nounwind {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* @llvm_mips_bclr_w_ARG1
+  %1 = load <4 x i32>, <4 x i32>* @llvm_mips_bclr_w_ARG2
+  %2 = tail call <4 x i32> @llvm.mips.bclr.w(<4 x i32> %0, <4 x i32> %1)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_bclr_w_RES
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.bclr.w(<4 x i32>, <4 x i32>) nounwind
+
+; CHECK-LABEL: llvm_mips_bclr_w_test:
+; CHECK-NOT: and.v
+; CHECK: bclr.w
+
+@llvm_mips_bclr_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_bclr_d_ARG2 = global <2 x i64> <i64 2, i64 3>, align 16
+@llvm_mips_bclr_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_bclr_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @llvm_mips_bclr_d_ARG1
+  %1 = load <2 x i64>, <2 x i64>* @llvm_mips_bclr_d_ARG2
+  %2 = tail call <2 x i64> @llvm.mips.bclr.d(<2 x i64> %0, <2 x i64> %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_bclr_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.bclr.d(<2 x i64>, <2 x i64>) nounwind
+
+; CHECK-LABEL: llvm_mips_bclr_d_test:
+; CHECK-NOT: and.v
+; CHECK: bclr.d
+
+@llvm_mips_bneg_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
+@llvm_mips_bneg_b_ARG2 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
+@llvm_mips_bneg_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_bneg_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_bneg_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_bneg_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.bneg.b(<16 x i8> %0, <16 x i8> %1)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_bneg_b_RES
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.bneg.b(<16 x i8>, <16 x i8>) nounwind
+
+; CHECK-LABEL: llvm_mips_bneg_b_test:
+; CHECK-NOT: andi.b
+; CHECK: bneg.b
+
+@llvm_mips_bneg_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
+@llvm_mips_bneg_h_ARG2 = global <8 x i16> <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, align 16
+@llvm_mips_bneg_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_bneg_h_test() nounwind {
+entry:
+  %0 = load <8 x i16>, <8 x i16>* @llvm_mips_bneg_h_ARG1
+  %1 = load <8 x i16>, <8 x i16>* @llvm_mips_bneg_h_ARG2
+  %2 = tail call <8 x i16> @llvm.mips.bneg.h(<8 x i16> %0, <8 x i16> %1)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_bneg_h_RES
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.bneg.h(<8 x i16>, <8 x i16>) nounwind
+
+; CHECK-LABEL: llvm_mips_bneg_h_test:
+; CHECK-NOT: and.v
+; CHECK: bneg.h
+
+@llvm_mips_bneg_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
+@llvm_mips_bneg_w_ARG2 = global <4 x i32> <i32 4, i32 5, i32 6, i32 7>, align 16
+@llvm_mips_bneg_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_bneg_w_test() nounwind {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* @llvm_mips_bneg_w_ARG1
+  %1 = load <4 x i32>, <4 x i32>* @llvm_mips_bneg_w_ARG2
+  %2 = tail call <4 x i32> @llvm.mips.bneg.w(<4 x i32> %0, <4 x i32> %1)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_bneg_w_RES
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.bneg.w(<4 x i32>, <4 x i32>) nounwind
+
+; CHECK-LABEL: llvm_mips_bneg_w_test:
+; CHECK-NOT: and.v
+; CHECK: bneg.w
+
+@llvm_mips_bneg_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_bneg_d_ARG2 = global <2 x i64> <i64 2, i64 3>, align 16
+@llvm_mips_bneg_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_bneg_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @llvm_mips_bneg_d_ARG1
+  %1 = load <2 x i64>, <2 x i64>* @llvm_mips_bneg_d_ARG2
+  %2 = tail call <2 x i64> @llvm.mips.bneg.d(<2 x i64> %0, <2 x i64> %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_bneg_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.bneg.d(<2 x i64>, <2 x i64>) nounwind
+
+; CHECK-LABEL: llvm_mips_bneg_d_test:
+; CHECK-NOT: and.v
+; CHECK: bneg.d
+
+@llvm_mips_bset_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
+@llvm_mips_bset_b_ARG2 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
+@llvm_mips_bset_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_bset_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_bset_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_bset_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.bset.b(<16 x i8> %0, <16 x i8> %1)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_bset_b_RES
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.bset.b(<16 x i8>, <16 x i8>) nounwind
+
+; CHECK-LABEL: llvm_mips_bset_b_test:
+; CHECK-NOT: andi.b
+; CHECK: bset.b
+
+@llvm_mips_bset_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
+@llvm_mips_bset_h_ARG2 = global <8 x i16> <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, align 16
+@llvm_mips_bset_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_bset_h_test() nounwind {
+entry:
+  %0 = load <8 x i16>, <8 x i16>* @llvm_mips_bset_h_ARG1
+  %1 = load <8 x i16>, <8 x i16>* @llvm_mips_bset_h_ARG2
+  %2 = tail call <8 x i16> @llvm.mips.bset.h(<8 x i16> %0, <8 x i16> %1)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_bset_h_RES
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.bset.h(<8 x i16>, <8 x i16>) nounwind
+
+; CHECK-LABEL: llvm_mips_bset_h_test:
+; CHECK-NOT: and.v
+; CHECK: bset.h
+
+@llvm_mips_bset_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
+@llvm_mips_bset_w_ARG2 = global <4 x i32> <i32 4, i32 5, i32 6, i32 7>, align 16
+@llvm_mips_bset_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_bset_w_test() nounwind {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* @llvm_mips_bset_w_ARG1
+  %1 = load <4 x i32>, <4 x i32>* @llvm_mips_bset_w_ARG2
+  %2 = tail call <4 x i32> @llvm.mips.bset.w(<4 x i32> %0, <4 x i32> %1)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_bset_w_RES
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.bset.w(<4 x i32>, <4 x i32>) nounwind
+
+; CHECK-LABEL: llvm_mips_bset_w_test:
+; CHECK-NOT: and.v
+; CHECK: bset.w
+
+@llvm_mips_bset_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_bset_d_ARG2 = global <2 x i64> <i64 2, i64 3>, align 16
+@llvm_mips_bset_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_bset_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @llvm_mips_bset_d_ARG1
+  %1 = load <2 x i64>, <2 x i64>* @llvm_mips_bset_d_ARG2
+  %2 = tail call <2 x i64> @llvm.mips.bset.d(<2 x i64> %0, <2 x i64> %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_bset_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.bset.d(<2 x i64>, <2 x i64>) nounwind
+
+; CHECK-LABEL: llvm_mips_bset_d_test:
+; CHECK-NOT: and.v
+; CHECK: bset.d
+
+@llvm_mips_sll_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
+@llvm_mips_sll_b_ARG2 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
+@llvm_mips_sll_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_sll_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_sll_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_sll_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.sll.b(<16 x i8> %0, <16 x i8> %1)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_sll_b_RES
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.sll.b(<16 x i8>, <16 x i8>) nounwind
+
+; CHECK-LABEL: llvm_mips_sll_b_test:
+; CHECK-NOT: andi.b
+; CHECK: sll.b
+
+@llvm_mips_sll_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
+@llvm_mips_sll_h_ARG2 = global <8 x i16> <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, align 16
+@llvm_mips_sll_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_sll_h_test() nounwind {
+entry:
+  %0 = load <8 x i16>, <8 x i16>* @llvm_mips_sll_h_ARG1
+  %1 = load <8 x i16>, <8 x i16>* @llvm_mips_sll_h_ARG2
+  %2 = tail call <8 x i16> @llvm.mips.sll.h(<8 x i16> %0, <8 x i16> %1)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_sll_h_RES
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.sll.h(<8 x i16>, <8 x i16>) nounwind
+
+; CHECK-LABEL: llvm_mips_sll_h_test:
+; CHECK-NOT: and.v
+; CHECK: sll.h
+
+@llvm_mips_sll_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
+@llvm_mips_sll_w_ARG2 = global <4 x i32> <i32 4, i32 5, i32 6, i32 7>, align 16
+@llvm_mips_sll_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_sll_w_test() nounwind {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* @llvm_mips_sll_w_ARG1
+  %1 = load <4 x i32>, <4 x i32>* @llvm_mips_sll_w_ARG2
+  %2 = tail call <4 x i32> @llvm.mips.sll.w(<4 x i32> %0, <4 x i32> %1)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_sll_w_RES
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.sll.w(<4 x i32>, <4 x i32>) nounwind
+
+; CHECK-LABEL: llvm_mips_sll_w_test:
+; CHECK-NOT: and.v
+; CHECK: sll.w
+
+@llvm_mips_sll_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_sll_d_ARG2 = global <2 x i64> <i64 2, i64 3>, align 16
+@llvm_mips_sll_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_sll_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @llvm_mips_sll_d_ARG1
+  %1 = load <2 x i64>, <2 x i64>* @llvm_mips_sll_d_ARG2
+  %2 = tail call <2 x i64> @llvm.mips.sll.d(<2 x i64> %0, <2 x i64> %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_sll_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.sll.d(<2 x i64>, <2 x i64>) nounwind
+
+; CHECK-LABEL: llvm_mips_sll_d_test:
+; CHECK-NOT: and.v
+; CHECK: sll.d
+
+@llvm_mips_sra_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
+@llvm_mips_sra_b_ARG2 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
+@llvm_mips_sra_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_sra_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_sra_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_sra_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.sra.b(<16 x i8> %0, <16 x i8> %1)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_sra_b_RES
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.sra.b(<16 x i8>, <16 x i8>) nounwind
+
+; CHECK-LABEL: llvm_mips_sra_b_test:
+; CHECK-NOT: andi.b
+; CHECK: sra.b
+
+@llvm_mips_sra_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
+@llvm_mips_sra_h_ARG2 = global <8 x i16> <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, align 16
+@llvm_mips_sra_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_sra_h_test() nounwind {
+entry:
+  %0 = load <8 x i16>, <8 x i16>* @llvm_mips_sra_h_ARG1
+  %1 = load <8 x i16>, <8 x i16>* @llvm_mips_sra_h_ARG2
+  %2 = tail call <8 x i16> @llvm.mips.sra.h(<8 x i16> %0, <8 x i16> %1)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_sra_h_RES
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.sra.h(<8 x i16>, <8 x i16>) nounwind
+
+; CHECK-LABEL: llvm_mips_sra_h_test:
+; CHECK-NOT: and.v
+; CHECK: sra.h
+
+@llvm_mips_sra_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
+@llvm_mips_sra_w_ARG2 = global <4 x i32> <i32 4, i32 5, i32 6, i32 7>, align 16
+@llvm_mips_sra_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_sra_w_test() nounwind {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* @llvm_mips_sra_w_ARG1
+  %1 = load <4 x i32>, <4 x i32>* @llvm_mips_sra_w_ARG2
+  %2 = tail call <4 x i32> @llvm.mips.sra.w(<4 x i32> %0, <4 x i32> %1)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_sra_w_RES
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.sra.w(<4 x i32>, <4 x i32>) nounwind
+
+; CHECK-LABEL: llvm_mips_sra_w_test:
+; CHECK-NOT: and.v
+; CHECK: sra.w
+
+@llvm_mips_sra_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_sra_d_ARG2 = global <2 x i64> <i64 2, i64 3>, align 16
+@llvm_mips_sra_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_sra_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @llvm_mips_sra_d_ARG1
+  %1 = load <2 x i64>, <2 x i64>* @llvm_mips_sra_d_ARG2
+  %2 = tail call <2 x i64> @llvm.mips.sra.d(<2 x i64> %0, <2 x i64> %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_sra_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.sra.d(<2 x i64>, <2 x i64>) nounwind
+
+; CHECK-LABEL: llvm_mips_sra_d_test:
+; CHECK-NOT: and.v
+; CHECK: sra.d
+
+@llvm_mips_srl_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
+@llvm_mips_srl_b_ARG2 = global <16 x i8> <i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31>, align 16
+@llvm_mips_srl_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_srl_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_srl_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_srl_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.srl.b(<16 x i8> %0, <16 x i8> %1)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_srl_b_RES
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.srl.b(<16 x i8>, <16 x i8>) nounwind
+
+; CHECK-LABEL: llvm_mips_srl_b_test:
+; CHECK-NOT: andi.b
+; CHECK: srl.b
+
+@llvm_mips_srl_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
+@llvm_mips_srl_h_ARG2 = global <8 x i16> <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, align 16
+@llvm_mips_srl_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_srl_h_test() nounwind {
+entry:
+  %0 = load <8 x i16>, <8 x i16>* @llvm_mips_srl_h_ARG1
+  %1 = load <8 x i16>, <8 x i16>* @llvm_mips_srl_h_ARG2
+  %2 = tail call <8 x i16> @llvm.mips.srl.h(<8 x i16> %0, <8 x i16> %1)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_srl_h_RES
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.srl.h(<8 x i16>, <8 x i16>) nounwind
+
+; CHECK-LABEL: llvm_mips_srl_h_test:
+; CHECK-NOT: and.v
+; CHECK: srl.h
+
+@llvm_mips_srl_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
+@llvm_mips_srl_w_ARG2 = global <4 x i32> <i32 4, i32 5, i32 6, i32 7>, align 16
+@llvm_mips_srl_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_srl_w_test() nounwind {
+entry:
+  %0 = load <4 x i32>, <4 x i32>* @llvm_mips_srl_w_ARG1
+  %1 = load <4 x i32>, <4 x i32>* @llvm_mips_srl_w_ARG2
+  %2 = tail call <4 x i32> @llvm.mips.srl.w(<4 x i32> %0, <4 x i32> %1)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_srl_w_RES
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.srl.w(<4 x i32>, <4 x i32>) nounwind
+
+; CHECK-LABEL: llvm_mips_srl_w_test:
+; CHECK-NOT: and.v
+; CHECK: srl.w
+
+@llvm_mips_srl_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
+@llvm_mips_srl_d_ARG2 = global <2 x i64> <i64 2, i64 3>, align 16
+@llvm_mips_srl_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_srl_d_test() nounwind {
+entry:
+  %0 = load <2 x i64>, <2 x i64>* @llvm_mips_srl_d_ARG1
+  %1 = load <2 x i64>, <2 x i64>* @llvm_mips_srl_d_ARG2
+  %2 = tail call <2 x i64> @llvm.mips.srl.d(<2 x i64> %0, <2 x i64> %1)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_srl_d_RES
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.srl.d(<2 x i64>, <2 x i64>) nounwind
+
+; CHECK-LABEL: llvm_mips_srl_d_test:
+; CHECK-NOT: and.v
+; CHECK: srl.d