AVX512 : VMOVSHDUP/VMOVSLDUP implementation.

[igorb]

Summary

AVX512: Implemented encoding and intrinsics for VMOVSHDUP/VMOVSLDUP instructions.
Added tests for intrinsics and encoding.

[delena]

lib/Target/X86/InstPrinter/X86InstComments.cpp

155

I propose to add some helper functions, like isZMMReg(), isYMMReg(), isXMMReg(),
it will simplify value type detection in some cases. And may be useful later.

191

We can examine the number of operands in order see if the instruction has memory operand.

225

remove empty line

[igorb]

Thanks you for review!

[delena]

lib/Target/X86/InstPrinter/X86InstComments.cpp

36

Operand index should be explicit.
The second parameter: MVT ScalarVT

379

missing comments

[RKSimon]

lib/Target/X86/InstPrinter/X86InstComments.cpp

379

Wouldn't it be better to match the 'fall through' pattern that we have for most rm / rr shuffle pairs instead of using operand count to re-determine if its rm? It would mean a splitting of the CASE_MOVDUP into 2 versions though.

[igorb]

Thanks for your comments and suggestions.

[RKSimon]

lib/Target/X86/InstPrinter/X86InstComments.cpp

173

We already have CASE_VSHUF_COMMON / CASE_VSHUF - is there any way that we can start standardizing these macros so that they can be reused as much as possible please?

[igorb]

Thanks for your comments and suggestions.

[RKSimon]

LGTM - Elena should confirm for the actual AVX512 encodings however.

[delena]

LGTM

[chapuni]

It broke layering violation. Reproducible with BUILD_SHARED_LIBS=ON.

X86AsmPrinter might require X86Desc but X86Desc requires X86AsmPrinter.

[igorb]

In D14322#289679, @chapuni wrote:

It broke layering violation. Reproducible with BUILD_SHARED_LIBS=ON.

X86AsmPrinter might require X86Desc but X86Desc requires X86AsmPrinter.

Thanks ,
Revert commit.

include/llvm/IR/IntrinsicsX86.td

def int_x86_avx512_mask_shuf_ps_256 :
           [llvm_v8f32_ty, llvm_v8f32_ty, llvm_i32_ty, llvm_v8f32_ty, llvm_i8_ty],
           [IntrNoMem]>;
 
   def int_x86_avx512_mask_shuf_ps_512 :
          GCCBuiltin<"__builtin_ia32_shufps512_mask">,
           Intrinsic<[llvm_v16f32_ty],
           [llvm_v16f32_ty, llvm_v16f32_ty, llvm_i32_ty, llvm_v16f32_ty, llvm_i16_ty],
           [IntrNoMem]>;
+
+  def int_x86_avx512_mask_movshdup_128 :
+         GCCBuiltin<"__builtin_ia32_movshdup128_mask">,
+          Intrinsic<[llvm_v4f32_ty],
+          [llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_movshdup_256 :
+         GCCBuiltin<"__builtin_ia32_movshdup256_mask">,
+          Intrinsic<[llvm_v8f32_ty],
+          [llvm_v8f32_ty, llvm_v8f32_ty, llvm_i8_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_movshdup_512 :
+         GCCBuiltin<"__builtin_ia32_movshdup512_mask">,
+          Intrinsic<[llvm_v16f32_ty],
+          [llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_movsldup_128 :
+         GCCBuiltin<"__builtin_ia32_movsldup128_mask">,
+          Intrinsic<[llvm_v4f32_ty],
+          [llvm_v4f32_ty, llvm_v4f32_ty, llvm_i8_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_movsldup_256 :
+         GCCBuiltin<"__builtin_ia32_movsldup256_mask">,
+          Intrinsic<[llvm_v8f32_ty],
+          [llvm_v8f32_ty, llvm_v8f32_ty, llvm_i8_ty],
+          [IntrNoMem]>;
+
+  def int_x86_avx512_mask_movsldup_512 :
+         GCCBuiltin<"__builtin_ia32_movsldup512_mask">,
+          Intrinsic<[llvm_v16f32_ty],
+          [llvm_v16f32_ty, llvm_v16f32_ty, llvm_i16_ty],
+          [IntrNoMem]>;
 }
 
 // Vector blend
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx_blendv_pd_256 : GCCBuiltin<"__builtin_ia32_blendvpd256">,
         Intrinsic<[llvm_v4f64_ty], [llvm_v4f64_ty,
                   llvm_v4f64_ty, llvm_v4f64_ty], [IntrNoMem]>;
   def int_x86_avx_blendv_ps_256 : GCCBuiltin<"__builtin_ia32_blendvps256">,

lib/Target/X86/InstPrinter/X86InstComments.cpp

 static void getZeroExtensionTypes(const MCInst *MI, MVT &SrcVT, MVT &DstVT) {
   switch (MI->getOpcode()) {
   default:
     llvm_unreachable("Unknown zero extension instruction");
   // i8 zero extension
   case X86::PMOVZXBWrm:
   case X86::PMOVZXBWrr:
   case X86::VPMOVZXBWrm:
   case X86::VPMOVZXBWrr:
     SrcVT = MVT::v16i8;
     DstVT = MVT::v8i16;
     break;
   case X86::VPMOVZXBWYrm:
   case X86::VPMOVZXBWYrr:
     SrcVT = MVT::v16i8;
     DstVT = MVT::v16i16;
     break;
static void getVSHUF64x2FamilyInfo(const MCInst *MI, MVT &VT, bool &HasMemOp) {
   CASE_VSHUF_COMMON(32X4, Z256, m)
     HasMemOp = true;        // FALL THROUGH.
   CASE_VSHUF_COMMON(32X4, Z256, r)
     VT = MVT::v8i32;
     break;
   }
 }
 
+#define CASE_MASK_INS_COMMON(Inst, Suffix, src)  \
+  case X86::V##Inst##Suffix##src:                \
+  case X86::V##Inst##Suffix##src##k:             \
+  case X86::V##Inst##Suffix##src##kz:
+
+#define CASE_SSE_INS_COMMON(Inst, src)  \
+  case X86::Inst##src:
+
+#define CASE_AVX_INS_COMMON(Inst, Suffix, src)  \
+  case X86::V##Inst##Suffix##src:
+
+#define CASE_MOVDUP(Inst)                \
+  CASE_MASK_INS_COMMON(Inst, Z, rr)      \
+  CASE_MASK_INS_COMMON(Inst, Z, rm)      \
+  CASE_MASK_INS_COMMON(Inst, Z256, rr)   \
+  CASE_MASK_INS_COMMON(Inst, Z256, rm)   \
+  CASE_MASK_INS_COMMON(Inst, Z128, rr)   \
+  CASE_MASK_INS_COMMON(Inst, Z128, rm)   \
+  CASE_AVX_INS_COMMON(Inst, , rr)        \
+  CASE_AVX_INS_COMMON(Inst, , rm)        \
+  CASE_AVX_INS_COMMON(Inst, Y, rr)       \
+  CASE_AVX_INS_COMMON(Inst, Y, rm)       \
+  CASE_SSE_INS_COMMON(Inst, rr)          \
+  CASE_SSE_INS_COMMON(Inst, rm)          \
+
+
+/// \brief Extracts the types and if it has memory operand for a given
+/// (MOVSHDUP/MOVSLDUP) instruction.
+static void getMOVSLDUPFamilyInfo(const MCInst *MI, MVT &VT, bool &HasMemOp) {
+  HasMemOp = false;
+  switch (MI->getOpcode()) {
+  default:
+    llvm_unreachable("Unknown MOVSHDUP family instructions.");
+    break;
+  CASE_MASK_INS_COMMON(MOVSHDUP, Z, rm)
+  CASE_MASK_INS_COMMON(MOVSLDUP, Z, rm)
+    HasMemOp = true;        // FALL THROUGH.
+  CASE_MASK_INS_COMMON(MOVSHDUP, Z, rr)
+  CASE_MASK_INS_COMMON(MOVSLDUP, Z, rr)
+    VT = MVT::v16f32;
+    break;
+  CASE_MASK_INS_COMMON(MOVSHDUP, Z256, rm)
+  CASE_MASK_INS_COMMON(MOVSLDUP, Z256, rm)
+  CASE_AVX_INS_COMMON(MOVSHDUP, Y, rm)
+  CASE_AVX_INS_COMMON(MOVSLDUP, Y, rm)
+    HasMemOp = true;        // FALL THROUGH.
+  CASE_MASK_INS_COMMON(MOVSHDUP, Z256, rr)
+  CASE_MASK_INS_COMMON(MOVSLDUP, Z256, rr)
+  CASE_AVX_INS_COMMON(MOVSHDUP, Y, rr)
+  CASE_AVX_INS_COMMON(MOVSLDUP, Y, rr)
+    VT = MVT::v8f32;
+    break;
+  CASE_MASK_INS_COMMON(MOVSHDUP, Z128, rm)
+  CASE_MASK_INS_COMMON(MOVSLDUP, Z128, rm)
+  CASE_AVX_INS_COMMON(MOVSHDUP, , rm)
+  CASE_AVX_INS_COMMON(MOVSLDUP, , rm)
+  CASE_SSE_INS_COMMON(MOVSHDUP, rm)
+  CASE_SSE_INS_COMMON(MOVSLDUP, rm)
+    HasMemOp = true;        // FALL THROUGH.
+  CASE_MASK_INS_COMMON(MOVSHDUP, Z128, rr)
+  CASE_MASK_INS_COMMON(MOVSLDUP, Z128, rr)
+  CASE_AVX_INS_COMMON(MOVSHDUP, , rr)
+  CASE_AVX_INS_COMMON(MOVSLDUP, , rr)
+  CASE_SSE_INS_COMMON(MOVSHDUP, rr)
+  CASE_SSE_INS_COMMON(MOVSLDUP, rr)
+    VT = MVT::v4f32;
+    break;
+  }
+}
+
+
 //===----------------------------------------------------------------------===//
 // Top Level Entrypoint
 //===----------------------------------------------------------------------===//
 
 /// EmitAnyX86InstComments - This function decodes x86 instructions and prints
 /// newline terminated strings to the specified string if desired.  This
 /// information is shown in disassembly dumps when verbose assembly is enabled.
 bool llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
bool llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
   case X86::MOVHLPSrr:
   case X86::VMOVHLPSrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     DecodeMOVHLPSMask(2, ShuffleMask);
     break;
 
-  case X86::MOVSLDUPrr:
-  case X86::VMOVSLDUPrr:
-    Src1Name = getRegName(MI->getOperand(1).getReg());
-    // FALL THROUGH.
-  case X86::MOVSLDUPrm:
-  case X86::VMOVSLDUPrm:
-    DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSLDUPMask(MVT::v4f32, ShuffleMask);
-    break;
-
-  case X86::VMOVSHDUPYrr:
+  CASE_MOVDUP(MOVSLDUP) {
+    MVT VT;
+    bool HasMemOp;
+    getMOVSLDUPFamilyInfo(MI, VT, HasMemOp);
+    if (!HasMemOp)
       Src1Name = getRegName(MI->getOperand(1).getReg());
-    // FALL THROUGH.
-  case X86::VMOVSHDUPYrm:
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSHDUPMask(MVT::v8f32, ShuffleMask);
+    DecodeMOVSLDUPMask(VT, ShuffleMask);
     break;
-
-  case X86::VMOVSLDUPYrr:
-    Src1Name = getRegName(MI->getOperand(1).getReg());
-    // FALL THROUGH.
-  case X86::VMOVSLDUPYrm:
-    DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSLDUPMask(MVT::v8f32, ShuffleMask);
-    break;
-
-  case X86::MOVSHDUPrr:
-  case X86::VMOVSHDUPrr:
+  }
+  CASE_MOVDUP(MOVSHDUP) {
+    MVT VT;
+    bool HasMemOp;
+    getMOVSLDUPFamilyInfo(MI, VT, HasMemOp);
+    if (!HasMemOp)
       Src1Name = getRegName(MI->getOperand(1).getReg());
-    // FALL THROUGH.
-  case X86::MOVSHDUPrm:
-  case X86::VMOVSHDUPrm:
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSHDUPMask(MVT::v4f32, ShuffleMask);
+    DecodeMOVSHDUPMask(VT, ShuffleMask);
     break;
-
+  }
   case X86::VMOVDDUPYrr:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     // FALL THROUGH.
   case X86::VMOVDDUPYrm:
     DestName = getRegName(MI->getOperand(0).getReg());
     DecodeMOVDDUPMask(MVT::v4f64, ShuffleMask);
     break;
 

lib/Target/X86/X86InstrAVX512.td

def rm  : AVX512PDI<0x12, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
                       (VT (X86Movddup (memop_frag addr:$src))))]>, EVEX;
 }
 
 defm VMOVDDUPZ : avx512_movddup<"vmovddup", VR512, v8f64, f512mem, loadv8f64>,
                  VEX_W, EVEX_V512, EVEX_CD8<64, CD8VF>;
 def : Pat<(X86Movddup (v8f64 (scalar_to_vector (loadf64 addr:$src)))),
           (VMOVDDUPZrm addr:$src)>;
 
-//===---------------------------------------------------------------------===//
-// Replicate Single FP - MOVSHDUP and MOVSLDUP
-//===---------------------------------------------------------------------===//
-multiclass avx512_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
-                              ValueType vt, RegisterClass RC, PatFrag mem_frag,
-                              X86MemOperand x86memop> {
-  def rr : AVX512XSI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
-                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                      [(set RC:$dst, (vt (OpNode RC:$src)))]>, EVEX;
-  let mayLoad = 1 in
-  def rm : AVX512XSI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
-                    !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
-                      [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>, EVEX;
-}
-
-defm VMOVSHDUPZ  : avx512_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
-                       v16f32, VR512, loadv16f32, f512mem>, EVEX_V512,
-                       EVEX_CD8<32, CD8VF>;
-defm VMOVSLDUPZ  : avx512_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
-                       v16f32, VR512, loadv16f32, f512mem>, EVEX_V512,
-                       EVEX_CD8<32, CD8VF>;
-
-def : Pat<(v16i32 (X86Movshdup VR512:$src)), (VMOVSHDUPZrr VR512:$src)>;
-def : Pat<(v16i32 (X86Movshdup (loadv16i32 addr:$src))),
-           (VMOVSHDUPZrm addr:$src)>;
-def : Pat<(v16i32 (X86Movsldup VR512:$src)), (VMOVSLDUPZrr VR512:$src)>;
-def : Pat<(v16i32 (X86Movsldup (loadv16i32 addr:$src))),
-           (VMOVSLDUPZrm addr:$src)>;
-
 //===----------------------------------------------------------------------===//
 // Move Low to High and High to Low packed FP Instructions
 //===----------------------------------------------------------------------===//
 def VMOVLHPSZrr : AVX512PSI<0x16, MRMSrcReg, (outs VR128X:$dst),
           (ins VR128X:$src1, VR128X:$src2),
           "vmovlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
           [(set VR128X:$dst, (v4f32 (X86Movlhps VR128X:$src1, VR128X:$src2)))],
            IIC_SSE_MOV_LH>, EVEX_4V;
defm VPALIGN:   avx512_common_3Op_rm_imm8<0x0F, X86PAlignr, "vpalignr" ,
                 EVEX_CD8<8, CD8VF>;
 
 defm VDBPSADBW: avx512_common_3Op_rm_imm8<0x42, X86dbpsadbw, "vdbpsadbw" ,
                     avx512vl_i16_info, avx512vl_i8_info>, EVEX_CD8<8, CD8VF>;
 
 multiclass avx512_unary_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
                            X86VectorVTInfo _> {
   defm rr : AVX512_maskable<opc, MRMSrcReg, _, (outs _.RC:$dst),
-                    (ins _.RC:$src1), OpcodeStr##_.Suffix,
+                    (ins _.RC:$src1), OpcodeStr,
                     "$src1", "$src1",
                     (_.VT (OpNode _.RC:$src1))>, EVEX, AVX5128IBase;
 
   let mayLoad = 1 in
     defm rm : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
-                    (ins _.MemOp:$src1), OpcodeStr##_.Suffix,
+                    (ins _.MemOp:$src1), OpcodeStr,
                     "$src1", "$src1",
                     (_.VT (OpNode (bitconvert (_.LdFrag addr:$src1))))>,
               EVEX, AVX5128IBase, EVEX_CD8<_.EltSize, CD8VF>;
 }
 
 multiclass avx512_unary_rmb<bits<8> opc, string OpcodeStr, SDNode OpNode,
                             X86VectorVTInfo _> :
            avx512_unary_rm<opc, OpcodeStr, OpNode, _> {
   let mayLoad = 1 in
     defm rmb : AVX512_maskable<opc, MRMSrcMem, _, (outs _.RC:$dst),
-                    (ins _.ScalarMemOp:$src1), OpcodeStr##_.Suffix,
+                    (ins _.ScalarMemOp:$src1), OpcodeStr,
                     "${src1}"##_.BroadcastStr,
                     "${src1}"##_.BroadcastStr,
                     (_.VT (OpNode (X86VBroadcast
                                       (_.ScalarLdFrag addr:$src1))))>,
                EVEX, AVX5128IBase, EVEX_B, EVEX_CD8<_.EltSize, CD8VF>;
 }
 
 multiclass avx512_unary_rm_vl<bits<8> opc, string OpcodeStr, SDNode OpNode,
defm Z256 : avx512_unary_rmb<opc, OpcodeStr, OpNode, VTInfo.info256>,
                                  EVEX_V256;
     defm Z128 : avx512_unary_rmb<opc, OpcodeStr, OpNode, VTInfo.info128>,
                                  EVEX_V128;
   }
 }
 
 multiclass avx512_unary_rm_vl_dq<bits<8> opc_d, bits<8> opc_q, string OpcodeStr,
                                  SDNode OpNode, Predicate prd> {
-  defm Q : avx512_unary_rmb_vl<opc_q, OpcodeStr, OpNode, avx512vl_i64_info,
+  defm Q : avx512_unary_rmb_vl<opc_q, OpcodeStr#"q", OpNode, avx512vl_i64_info,
                                prd>, VEX_W;
-  defm D : avx512_unary_rmb_vl<opc_d, OpcodeStr, OpNode, avx512vl_i32_info, prd>;
+  defm D : avx512_unary_rmb_vl<opc_d, OpcodeStr#"d", OpNode, avx512vl_i32_info,
+                               prd>;
 }
 
 multiclass avx512_unary_rm_vl_bw<bits<8> opc_b, bits<8> opc_w, string OpcodeStr,
                                  SDNode OpNode, Predicate prd> {
-  defm W : avx512_unary_rm_vl<opc_w, OpcodeStr, OpNode, avx512vl_i16_info, prd>;
-  defm B : avx512_unary_rm_vl<opc_b, OpcodeStr, OpNode, avx512vl_i8_info, prd>;
+  defm W : avx512_unary_rm_vl<opc_w, OpcodeStr#"w", OpNode, avx512vl_i16_info, prd>;
+  defm B : avx512_unary_rm_vl<opc_b, OpcodeStr#"b", OpNode, avx512vl_i8_info, prd>;
 }
 
 multiclass avx512_unary_rm_vl_all<bits<8> opc_b, bits<8> opc_w,
                                   bits<8> opc_d, bits<8> opc_q,
                                   string OpcodeStr, SDNode OpNode> {
   defm NAME : avx512_unary_rm_vl_dq<opc_d, opc_q, OpcodeStr, OpNode,
                                     HasAVX512>,
               avx512_unary_rm_vl_bw<opc_b, opc_w, OpcodeStr, OpNode,
multiclass avx512_ctlz<bits<8> opc, string OpcodeStr, Predicate prd>{
   let isCodeGenOnly = 1 in
     defm NAME#_UNDEF : avx512_unary_rm_vl_dq<opc, opc, OpcodeStr,
                                              ctlz_zero_undef, prd>;
 }
 
 defm VPLZCNT    : avx512_ctlz<0x44, "vplzcnt", HasCDI>;
 defm VPCONFLICT : avx512_unary_rm_vl_dq<0xC4, 0xC4, "vpconflict", X86Conflict, HasCDI>;
 
+//===---------------------------------------------------------------------===//
+// Replicate Single FP - MOVSHDUP and MOVSLDUP
+//===---------------------------------------------------------------------===//
+multiclass avx512_replicate<bits<8> opc, string OpcodeStr, SDNode OpNode>{
+  defm NAME:       avx512_unary_rm_vl<opc, OpcodeStr, OpNode, avx512vl_f32_info,
+                                      HasAVX512>, XS;
+  let isCodeGenOnly = 1 in
+    defm NAME#_I: avx512_unary_rm_vl<opc, OpcodeStr, OpNode, avx512vl_i32_info,
+                                     HasAVX512>, XS;
+}
+
+defm VMOVSHDUP : avx512_replicate<0x16, "vmovshdup", X86Movshdup>;
+defm VMOVSLDUP : avx512_replicate<0x12, "vmovsldup", X86Movsldup>;
 //===----------------------------------------------------------------------===//
 // AVX-512 - Unpack Instructions
 //===----------------------------------------------------------------------===//
 defm VUNPCKH : avx512_fp_binop_p<0x15, "vunpckh", X86Unpckh>;
 defm VUNPCKL : avx512_fp_binop_p<0x14, "vunpckl", X86Unpckl>;
 
 defm VPUNPCKLBW : avx512_binop_rm_vl_b<0x60, "vpunpcklbw", X86Unpckl,
                                        SSE_INTALU_ITINS_P, HasBWI>;

lib/Target/X86/X86InstrSSE.td

def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
                       [(set RC:$dst, (vt (OpNode RC:$src)))],
                       IIC_SSE_MOV_LH>, Sched<[WriteFShuffle]>;
 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
                     !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
                       [(set RC:$dst, (OpNode (mem_frag addr:$src)))],
                       IIC_SSE_MOV_LH>, Sched<[WriteLoad]>;
 }
 
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
   defm VMOVSHDUP  : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
                                        v4f32, VR128, loadv4f32, f128mem>, VEX;
   defm VMOVSLDUP  : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
                                        v4f32, VR128, loadv4f32, f128mem>, VEX;
   defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
                                  v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
   defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
                                  v8f32, VR256, loadv8f32, f256mem>, VEX, VEX_L;
 }
 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
                                    memopv4f32, f128mem>;
 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
                                    memopv4f32, f128mem>;
 
-let Predicates = [HasAVX] in {
+let Predicates = [HasAVX, NoVLX] in {
   def : Pat<(v4i32 (X86Movshdup VR128:$src)),
             (VMOVSHDUPrr VR128:$src)>;
   def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (loadv2i64 addr:$src)))),
             (VMOVSHDUPrm addr:$src)>;
   def : Pat<(v4i32 (X86Movsldup VR128:$src)),
             (VMOVSLDUPrr VR128:$src)>;
   def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (loadv2i64 addr:$src)))),
             (VMOVSLDUPrm addr:$src)>;

lib/Target/X86/X86IntrinsicsInfo.h

static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_max_pd_128, INTR_TYPE_2OP_MASK, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_pd_256, INTR_TYPE_2OP_MASK, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_pd_512, INTR_TYPE_2OP_MASK, X86ISD::FMAX,
                      X86ISD::FMAX_RND),
   X86_INTRINSIC_DATA(avx512_mask_max_ps_128, INTR_TYPE_2OP_MASK, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_ps_256, INTR_TYPE_2OP_MASK, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(avx512_mask_max_ps_512, INTR_TYPE_2OP_MASK, X86ISD::FMAX,
                      X86ISD::FMAX_RND),
-  X86_INTRINSIC_DATA(avx512_mask_max_sd_round, INTR_TYPE_SCALAR_MASK_RM, X86ISD::FMAX,
-  X86ISD::FMAX_RND),
-  X86_INTRINSIC_DATA(avx512_mask_max_ss_round, INTR_TYPE_SCALAR_MASK_RM, X86ISD::FMAX,
-  X86ISD::FMAX_RND),
+  X86_INTRINSIC_DATA(avx512_mask_max_sd_round, INTR_TYPE_SCALAR_MASK_RM,
+                     X86ISD::FMAX, X86ISD::FMAX_RND),
+  X86_INTRINSIC_DATA(avx512_mask_max_ss_round, INTR_TYPE_SCALAR_MASK_RM,
+                     X86ISD::FMAX, X86ISD::FMAX_RND),
   X86_INTRINSIC_DATA(avx512_mask_min_pd_128, INTR_TYPE_2OP_MASK, X86ISD::FMIN, 0),
   X86_INTRINSIC_DATA(avx512_mask_min_pd_256, INTR_TYPE_2OP_MASK, X86ISD::FMIN, 0),
   X86_INTRINSIC_DATA(avx512_mask_min_pd_512, INTR_TYPE_2OP_MASK, X86ISD::FMIN,
                      X86ISD::FMIN_RND),
   X86_INTRINSIC_DATA(avx512_mask_min_ps_128, INTR_TYPE_2OP_MASK, X86ISD::FMIN, 0),
   X86_INTRINSIC_DATA(avx512_mask_min_ps_256, INTR_TYPE_2OP_MASK, X86ISD::FMIN, 0),
   X86_INTRINSIC_DATA(avx512_mask_min_ps_512, INTR_TYPE_2OP_MASK, X86ISD::FMIN,
                      X86ISD::FMIN_RND),
-  X86_INTRINSIC_DATA(avx512_mask_min_sd_round, INTR_TYPE_SCALAR_MASK_RM, X86ISD::FMIN,
-  X86ISD::FMIN_RND),
-  X86_INTRINSIC_DATA(avx512_mask_min_ss_round, INTR_TYPE_SCALAR_MASK_RM, X86ISD::FMIN,
-  X86ISD::FMIN_RND),
+  X86_INTRINSIC_DATA(avx512_mask_min_sd_round, INTR_TYPE_SCALAR_MASK_RM,
+                     X86ISD::FMIN, X86ISD::FMIN_RND),
+  X86_INTRINSIC_DATA(avx512_mask_min_ss_round, INTR_TYPE_SCALAR_MASK_RM,
+                     X86ISD::FMIN, X86ISD::FMIN_RND),
+  X86_INTRINSIC_DATA(avx512_mask_movshdup_128, INTR_TYPE_1OP_MASK,
+                     X86ISD::MOVSHDUP, 0),
+  X86_INTRINSIC_DATA(avx512_mask_movshdup_256, INTR_TYPE_1OP_MASK,
+                     X86ISD::MOVSHDUP, 0),
+  X86_INTRINSIC_DATA(avx512_mask_movshdup_512, INTR_TYPE_1OP_MASK,
+                     X86ISD::MOVSHDUP, 0),
+  X86_INTRINSIC_DATA(avx512_mask_movsldup_128, INTR_TYPE_1OP_MASK,
+                     X86ISD::MOVSLDUP, 0),
+  X86_INTRINSIC_DATA(avx512_mask_movsldup_256, INTR_TYPE_1OP_MASK,
+                     X86ISD::MOVSLDUP, 0),
+  X86_INTRINSIC_DATA(avx512_mask_movsldup_512, INTR_TYPE_1OP_MASK,
+                     X86ISD::MOVSLDUP, 0),
   X86_INTRINSIC_DATA(avx512_mask_mul_pd_128, INTR_TYPE_2OP_MASK, ISD::FMUL, 0),
   X86_INTRINSIC_DATA(avx512_mask_mul_pd_256, INTR_TYPE_2OP_MASK, ISD::FMUL, 0),
   X86_INTRINSIC_DATA(avx512_mask_mul_pd_512, INTR_TYPE_2OP_MASK, ISD::FMUL,
                      X86ISD::FMUL_RND),
   X86_INTRINSIC_DATA(avx512_mask_mul_ps_128, INTR_TYPE_2OP_MASK, ISD::FMUL, 0),
   X86_INTRINSIC_DATA(avx512_mask_mul_ps_256, INTR_TYPE_2OP_MASK, ISD::FMUL, 0),
   X86_INTRINSIC_DATA(avx512_mask_mul_ps_512, INTR_TYPE_2OP_MASK, ISD::FMUL,
                      X86ISD::FMUL_RND),

test/CodeGen/X86/avx-isa-check.ll

define <16 x i8> @shuffle_v16i8_0101010101010101(<16 x i8> %a, <16 x i8> %b) {
   ret <16 x i8> %shuffle
 }
 
 define <16 x i16> @shuffle_v16i16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00(<16 x i16> %a, <16 x i16> %b) {
   %shuffle = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <16 x i16> %shuffle
 }
 
+define <8 x float> @shuffle_v8f32_11335577(<8 x float> %a, <8 x float> %b) {
+; vmovshdup 256 test
+  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
+  ret <8 x float> %shuffle
+}
+
+define <4 x float> @shuffle_v4f32_1133(<4 x float> %a, <4 x float> %b) {
+; vmovshdup 128 test 
+  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
+  ret <4 x float> %shuffle
+}
+
+define <8 x float> @shuffle_v8f32_00224466(<8 x float> %a, <8 x float> %b) {
+; vmovsldup 256 test
+  %shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>
+  ret <8 x float> %shuffle
+}
+
+define <4 x float> @shuffle_v4f32_0022(<4 x float> %a, <4 x float> %b) {
+; vmovsldup 128 test
+  %shuffle = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
+  ret <4 x float> %shuffle
+}

test/CodeGen/X86/avx512-intrinsics.ll

 ; CHECK-NEXT:    vpaddq %zmm0, %zmm3, %zmm0
 ; CHECK-NEXT:    retq
   %res = call <8 x i64> @llvm.x86.avx512.maskz.pternlog.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i32 33, i8 %x4)
   %res1 = call <8 x i64> @llvm.x86.avx512.maskz.pternlog.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i32 33, i8 -1)
   %res2 = add <8 x i64> %res, %res1
   ret <8 x i64> %res2
 }
 
+declare <16 x float> @llvm.x86.avx512.mask.movsldup.512(<16 x float>, <16 x float>, i16)
+
+define <16 x float>@test_int_x86_avx512_mask_movsldup_512(<16 x float> %x0, <16 x float> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_movsldup_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1 
+; CHECK-NEXT:    vmovsldup %zmm0, %zmm1 {%k1} 
+; CHECK-NEXT:    ## zmm1 = zmm1[0,0,2,2,4,4,6,6,8,8,10,10,12,12,14,14]
+; CHECK-NEXT:    vmovsldup %zmm0, %zmm2 {%k1} {z} 
+; CHECK-NEXT:    ## zmm2 = k1[0,0,2,2,4,4,6,6,8,8,10,10,12,12,14,14]
+; CHECK-NEXT:    vmovsldup %zmm0, %zmm0 
+; CHECK-NEXT:    ## zmm0 = zmm0[0,0,2,2,4,4,6,6,8,8,10,10,12,12,14,14]
+; CHECK-NEXT:    vaddps %zmm0, %zmm1, %zmm0 
+; CHECK-NEXT:    vaddps %zmm0, %zmm2, %zmm0 
+; CHECK-NEXT:    retq 
+  %res = call <16 x float> @llvm.x86.avx512.mask.movsldup.512(<16 x float> %x0, <16 x float> %x1, i16 %x2)
+  %res1 = call <16 x float> @llvm.x86.avx512.mask.movsldup.512(<16 x float> %x0, <16 x float> %x1, i16 -1)
+  %res2 = call <16 x float> @llvm.x86.avx512.mask.movsldup.512(<16 x float> %x0, <16 x float> zeroinitializer, i16 %x2)
+  %res3 = fadd <16 x float> %res, %res1
+  %res4 = fadd <16 x float> %res2, %res3
+  ret <16 x float> %res4
+}
+
+declare <16 x float> @llvm.x86.avx512.mask.movshdup.512(<16 x float>, <16 x float>, i16)
+
+define <16 x float>@test_int_x86_avx512_mask_movshdup_512(<16 x float> %x0, <16 x float> %x1, i16 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_movshdup_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1 
+; CHECK-NEXT:    vmovshdup %zmm0, %zmm1 {%k1} 
+; CHECK-NEXT:    ## zmm1 = zmm1[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; CHECK-NEXT:    vmovshdup %zmm0, %zmm2 {%k1} {z} 
+; CHECK-NEXT:    ## zmm2 = k1[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; CHECK-NEXT:    vmovshdup %zmm0, %zmm0 
+; CHECK-NEXT:    ## zmm0 = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
+; CHECK-NEXT:    vaddps %zmm0, %zmm1, %zmm0 
+; CHECK-NEXT:    vaddps %zmm0, %zmm2, %zmm0 
+; CHECK-NEXT:    retq 
+  %res = call <16 x float> @llvm.x86.avx512.mask.movshdup.512(<16 x float> %x0, <16 x float> %x1, i16 %x2)
+  %res1 = call <16 x float> @llvm.x86.avx512.mask.movshdup.512(<16 x float> %x0, <16 x float> %x1, i16 -1)
+  %res2 = call <16 x float> @llvm.x86.avx512.mask.movshdup.512(<16 x float> %x0, <16 x float> zeroinitializer, i16 %x2)
+  %res3 = fadd <16 x float> %res, %res1
+  %res4 = fadd <16 x float> %res2, %res3
+  ret <16 x float> %res4
+}
+

test/CodeGen/X86/avx512vl-intrinsics.ll

 define <8 x i16> @test_x86_vcvtps2ph_256(<8 x float> %a0) {
   ; CHECK: test_x86_vcvtps2ph_256
   ; CHECK: vcvtps2ph $2, %ymm0, %xmm0
   %res = call <8 x i16> @llvm.x86.avx512.mask.vcvtps2ph.256(<8 x float> %a0, i32 2, <8 x i16> zeroinitializer, i8 -1)
   ret <8 x i16> %res
 }
 
 declare <8 x i16> @llvm.x86.avx512.mask.vcvtps2ph.256(<8 x float>, i32, <8 x i16>, i8) nounwind readonly
+
+declare <4 x float> @llvm.x86.avx512.mask.movsldup.128(<4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask_movsldup_128(<4 x float> %x0, <4 x float> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_movsldup_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax 
+; CHECK-NEXT:    kmovw %eax, %k1 
+; CHECK-NEXT:    vmovsldup %xmm0, %xmm1 {%k1} 
+; CHECK-NEXT:    ## xmm1 = xmm1[0,0,2,2]
+; CHECK-NEXT:    vmovsldup %xmm0, %xmm2 {%k1} {z} 
+; CHECK-NEXT:    ## xmm2 = k1[0,0,2,2]
+; CHECK-NEXT:    vmovsldup %xmm0, %xmm0 
+; CHECK-NEXT:    ## xmm0 = xmm0[0,0,2,2]
+; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0 
+; CHECK-NEXT:    vaddps %xmm0, %xmm2, %xmm0 
+; CHECK-NEXT:    retq 
+  %res = call <4 x float> @llvm.x86.avx512.mask.movsldup.128(<4 x float> %x0, <4 x float> %x1, i8 %x2)
+  %res1 = call <4 x float> @llvm.x86.avx512.mask.movsldup.128(<4 x float> %x0, <4 x float> %x1, i8 -1)
+  %res2 = call <4 x float> @llvm.x86.avx512.mask.movsldup.128(<4 x float> %x0, <4 x float> zeroinitializer, i8 %x2)
+  %res3 = fadd <4 x float> %res, %res1
+  %res4 = fadd <4 x float> %res2, %res3
+  ret <4 x float> %res4
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.movsldup.256(<8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask_movsldup_256(<8 x float> %x0, <8 x float> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_movsldup_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax 
+; CHECK-NEXT:    kmovw %eax, %k1 
+; CHECK-NEXT:    vmovsldup %ymm0, %ymm1 {%k1} 
+; CHECK-NEXT:    ## ymm1 = ymm1[0,0,2,2,4,4,6,6]
+; CHECK-NEXT:    vmovsldup %ymm0, %ymm2 {%k1} {z} 
+; CHECK-NEXT:    ## ymm2 = k1[0,0,2,2,4,4,6,6]
+; CHECK-NEXT:    vmovsldup %ymm0, %ymm0 
+; CHECK-NEXT:    ## ymm0 = ymm0[0,0,2,2,4,4,6,6]
+; CHECK-NEXT:    vaddps %ymm0, %ymm1, %ymm0 
+; CHECK-NEXT:    vaddps %ymm0, %ymm2, %ymm0 
+; CHECK-NEXT:    retq 
+  %res = call <8 x float> @llvm.x86.avx512.mask.movsldup.256(<8 x float> %x0, <8 x float> %x1, i8 %x2)
+  %res1 = call <8 x float> @llvm.x86.avx512.mask.movsldup.256(<8 x float> %x0, <8 x float> %x1, i8 -1)
+  %res2 = call <8 x float> @llvm.x86.avx512.mask.movsldup.256(<8 x float> %x0, <8 x float> zeroinitializer, i8 %x2)
+  %res3 = fadd <8 x float> %res, %res1
+  %res4 = fadd <8 x float> %res2, %res3
+  ret <8 x float> %res4
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.movshdup.128(<4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask_movshdup_128(<4 x float> %x0, <4 x float> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_movshdup_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax 
+; CHECK-NEXT:    kmovw %eax, %k1 
+; CHECK-NEXT:    vmovshdup %xmm0, %xmm1 {%k1} 
+; CHECK-NEXT:    ## xmm1 = xmm1[1,1,3,3]
+; CHECK-NEXT:    vmovshdup %xmm0, %xmm2 {%k1} {z} 
+; CHECK-NEXT:    ## xmm2 = k1[1,1,3,3]
+; CHECK-NEXT:    vmovshdup %xmm0, %xmm0 
+; CHECK-NEXT:    ## xmm0 = xmm0[1,1,3,3]
+; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0 
+; CHECK-NEXT:    vaddps %xmm0, %xmm2, %xmm0 
+; CHECK-NEXT:    retq 
+  %res = call <4 x float> @llvm.x86.avx512.mask.movshdup.128(<4 x float> %x0, <4 x float> %x1, i8 %x2)
+  %res1 = call <4 x float> @llvm.x86.avx512.mask.movshdup.128(<4 x float> %x0, <4 x float> %x1, i8 -1)
+  %res2 = call <4 x float> @llvm.x86.avx512.mask.movshdup.128(<4 x float> %x0, <4 x float> zeroinitializer, i8 %x2)
+  %res3 = fadd <4 x float> %res, %res1
+  %res4 = fadd <4 x float> %res2, %res3
+  ret <4 x float> %res4
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.movshdup.256(<8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask_movshdup_256(<8 x float> %x0, <8 x float> %x1, i8 %x2) {
+; CHECK-LABEL: test_int_x86_avx512_mask_movshdup_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax 
+; CHECK-NEXT:    kmovw %eax, %k1 
+; CHECK-NEXT:    vmovshdup %ymm0, %ymm1 {%k1} 
+; CHECK-NEXT:    ## ymm1 = ymm1[1,1,3,3,5,5,7,7]
+; CHECK-NEXT:    vmovshdup %ymm0, %ymm2 {%k1} {z} 
+; CHECK-NEXT:    ## ymm2 = k1[1,1,3,3,5,5,7,7]
+; CHECK-NEXT:    vmovshdup %ymm0, %ymm0 
+; CHECK-NEXT:    ## ymm0 = ymm0[1,1,3,3,5,5,7,7]
+; CHECK-NEXT:    vaddps %ymm0, %ymm1, %ymm0 
+; CHECK-NEXT:    vaddps %ymm0, %ymm2, %ymm0 
+; CHECK-NEXT:    retq 
+  %res = call <8 x float> @llvm.x86.avx512.mask.movshdup.256(<8 x float> %x0, <8 x float> %x1, i8 %x2)
+  %res1 = call <8 x float> @llvm.x86.avx512.mask.movshdup.256(<8 x float> %x0, <8 x float> %x1, i8 -1)
+  %res2 = call <8 x float> @llvm.x86.avx512.mask.movshdup.256(<8 x float> %x0, <8 x float> zeroinitializer, i8 %x2)
+  %res3 = fadd <8 x float> %res, %res1
+  %res4 = fadd <8 x float> %res2, %res3
+  ret <8 x float> %res4
+}
+

test/MC/X86/avx512-encodings.s

 // CHECK: vmovd  %xmm5, -512(%rdx)
 // CHECK:  encoding: [0xc5,0xf9,0x7e,0xaa,0x00,0xfe,0xff,0xff]
           vmovd  %xmm5, -512(%rdx)
 
 // CHECK: vmovd  %xmm5, -516(%rdx)
 // CHECK:  encoding: [0xc5,0xf9,0x7e,0xaa,0xfc,0xfd,0xff,0xff]
           vmovd  %xmm5, -516(%rdx)
 
+// CHECK: vmovshdup %zmm27, %zmm16
+// CHECK:  encoding: [0x62,0x81,0x7e,0x48,0x16,0xc3]
+          vmovshdup %zmm27, %zmm16
+
+// CHECK: vmovshdup %zmm27, %zmm16 {%k4}
+// CHECK:  encoding: [0x62,0x81,0x7e,0x4c,0x16,0xc3]
+          vmovshdup %zmm27, %zmm16 {%k4}
+
+// CHECK: vmovshdup %zmm27, %zmm16 {%k4} {z}
+// CHECK:  encoding: [0x62,0x81,0x7e,0xcc,0x16,0xc3]
+          vmovshdup %zmm27, %zmm16 {%k4} {z}
+
+// CHECK: vmovshdup (%rcx), %zmm16
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x48,0x16,0x01]
+          vmovshdup (%rcx), %zmm16
+
+// CHECK: vmovshdup 291(%rax,%r14,8), %zmm16
+// CHECK:  encoding: [0x62,0xa1,0x7e,0x48,0x16,0x84,0xf0,0x23,0x01,0x00,0x00]
+          vmovshdup 291(%rax,%r14,8), %zmm16
+
+// CHECK: vmovshdup 8128(%rdx), %zmm16
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x48,0x16,0x42,0x7f]
+          vmovshdup 8128(%rdx), %zmm16
+
+// CHECK: vmovshdup 8192(%rdx), %zmm16
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x48,0x16,0x82,0x00,0x20,0x00,0x00]
+          vmovshdup 8192(%rdx), %zmm16
+
+// CHECK: vmovshdup -8192(%rdx), %zmm16
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x48,0x16,0x42,0x80]
+          vmovshdup -8192(%rdx), %zmm16
+
+// CHECK: vmovshdup -8256(%rdx), %zmm16
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x48,0x16,0x82,0xc0,0xdf,0xff,0xff]
+          vmovshdup -8256(%rdx), %zmm16
+
+// CHECK: vmovsldup %zmm14, %zmm13
+// CHECK:  encoding: [0x62,0x51,0x7e,0x48,0x12,0xee]
+          vmovsldup %zmm14, %zmm13
+
+// CHECK: vmovsldup %zmm14, %zmm13 {%k6}
+// CHECK:  encoding: [0x62,0x51,0x7e,0x4e,0x12,0xee]
+          vmovsldup %zmm14, %zmm13 {%k6}
+
+// CHECK: vmovsldup %zmm14, %zmm13 {%k6} {z}
+// CHECK:  encoding: [0x62,0x51,0x7e,0xce,0x12,0xee]
+          vmovsldup %zmm14, %zmm13 {%k6} {z}
+
+// CHECK: vmovsldup (%rcx), %zmm13
+// CHECK:  encoding: [0x62,0x71,0x7e,0x48,0x12,0x29]
+          vmovsldup (%rcx), %zmm13
+
+// CHECK: vmovsldup 291(%rax,%r14,8), %zmm13
+// CHECK:  encoding: [0x62,0x31,0x7e,0x48,0x12,0xac,0xf0,0x23,0x01,0x00,0x00]
+          vmovsldup 291(%rax,%r14,8), %zmm13
+
+// CHECK: vmovsldup 8128(%rdx), %zmm13
+// CHECK:  encoding: [0x62,0x71,0x7e,0x48,0x12,0x6a,0x7f]
+          vmovsldup 8128(%rdx), %zmm13
+
+// CHECK: vmovsldup 8192(%rdx), %zmm13
+// CHECK:  encoding: [0x62,0x71,0x7e,0x48,0x12,0xaa,0x00,0x20,0x00,0x00]
+          vmovsldup 8192(%rdx), %zmm13
+
+// CHECK: vmovsldup -8192(%rdx), %zmm13
+// CHECK:  encoding: [0x62,0x71,0x7e,0x48,0x12,0x6a,0x80]
+          vmovsldup -8192(%rdx), %zmm13
+
+// CHECK: vmovsldup -8256(%rdx), %zmm13
+// CHECK:  encoding: [0x62,0x71,0x7e,0x48,0x12,0xaa,0xc0,0xdf,0xff,0xff]
+          vmovsldup -8256(%rdx), %zmm13
+

test/MC/X86/x86-64-avx512f_vl.s

 
 // CHECK: vcvtps2ph $123, %ymm30, -2048(%rdx)
 // CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0x72,0x80,0x7b]
           vcvtps2ph $0x7b, %ymm30, -2048(%rdx)
 
 // CHECK: vcvtps2ph $123, %ymm30, -2064(%rdx)
 // CHECK:  encoding: [0x62,0x63,0x7d,0x28,0x1d,0xb2,0xf0,0xf7,0xff,0xff,0x7b]
           vcvtps2ph $0x7b, %ymm30, -2064(%rdx)
+
+// CHECK: vmovshdup %xmm18, %xmm23
+// CHECK:  encoding: [0x62,0xa1,0x7e,0x08,0x16,0xfa]
+          vmovshdup %xmm18, %xmm23
+
+// CHECK: vmovshdup %xmm18, %xmm23 {%k2}
+// CHECK:  encoding: [0x62,0xa1,0x7e,0x0a,0x16,0xfa]
+          vmovshdup %xmm18, %xmm23 {%k2}
+
+// CHECK: vmovshdup %xmm18, %xmm23 {%k2} {z}
+// CHECK:  encoding: [0x62,0xa1,0x7e,0x8a,0x16,0xfa]
+          vmovshdup %xmm18, %xmm23 {%k2} {z}
+
+// CHECK: vmovshdup (%rcx), %xmm23
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x08,0x16,0x39]
+          vmovshdup (%rcx), %xmm23
+
+// CHECK: vmovshdup 291(%rax,%r14,8), %xmm23
+// CHECK:  encoding: [0x62,0xa1,0x7e,0x08,0x16,0xbc,0xf0,0x23,0x01,0x00,0x00]
+          vmovshdup 291(%rax,%r14,8), %xmm23
+
+// CHECK: vmovshdup 2032(%rdx), %xmm23
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x08,0x16,0x7a,0x7f]
+          vmovshdup 2032(%rdx), %xmm23
+
+// CHECK: vmovshdup 2048(%rdx), %xmm23
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x08,0x16,0xba,0x00,0x08,0x00,0x00]
+          vmovshdup 2048(%rdx), %xmm23
+
+// CHECK: vmovshdup -2048(%rdx), %xmm23
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x08,0x16,0x7a,0x80]
+          vmovshdup -2048(%rdx), %xmm23
+
+// CHECK: vmovshdup -2064(%rdx), %xmm23
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x08,0x16,0xba,0xf0,0xf7,0xff,0xff]
+          vmovshdup -2064(%rdx), %xmm23
+
+// CHECK: vmovshdup %ymm24, %ymm18
+// CHECK:  encoding: [0x62,0x81,0x7e,0x28,0x16,0xd0]
+          vmovshdup %ymm24, %ymm18
+
+// CHECK: vmovshdup %ymm24, %ymm18 {%k3}
+// CHECK:  encoding: [0x62,0x81,0x7e,0x2b,0x16,0xd0]
+          vmovshdup %ymm24, %ymm18 {%k3}
+
+// CHECK: vmovshdup %ymm24, %ymm18 {%k3} {z}
+// CHECK:  encoding: [0x62,0x81,0x7e,0xab,0x16,0xd0]
+          vmovshdup %ymm24, %ymm18 {%k3} {z}
+
+// CHECK: vmovshdup (%rcx), %ymm18
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x28,0x16,0x11]
+          vmovshdup (%rcx), %ymm18
+
+// CHECK: vmovshdup 291(%rax,%r14,8), %ymm18
+// CHECK:  encoding: [0x62,0xa1,0x7e,0x28,0x16,0x94,0xf0,0x23,0x01,0x00,0x00]
+          vmovshdup 291(%rax,%r14,8), %ymm18
+
+// CHECK: vmovshdup 4064(%rdx), %ymm18
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x28,0x16,0x52,0x7f]
+          vmovshdup 4064(%rdx), %ymm18
+
+// CHECK: vmovshdup 4096(%rdx), %ymm18
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x28,0x16,0x92,0x00,0x10,0x00,0x00]
+          vmovshdup 4096(%rdx), %ymm18
+
+// CHECK: vmovshdup -4096(%rdx), %ymm18
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x28,0x16,0x52,0x80]
+          vmovshdup -4096(%rdx), %ymm18
+
+// CHECK: vmovshdup -4128(%rdx), %ymm18
+// CHECK:  encoding: [0x62,0xe1,0x7e,0x28,0x16,0x92,0xe0,0xef,0xff,0xff]
+          vmovshdup -4128(%rdx), %ymm18
+
+// CHECK: vmovsldup %xmm21, %xmm25
+// CHECK:  encoding: [0x62,0x21,0x7e,0x08,0x12,0xcd]
+          vmovsldup %xmm21, %xmm25
+
+// CHECK: vmovsldup %xmm21, %xmm25 {%k5}
+// CHECK:  encoding: [0x62,0x21,0x7e,0x0d,0x12,0xcd]
+          vmovsldup %xmm21, %xmm25 {%k5}
+
+// CHECK: vmovsldup %xmm21, %xmm25 {%k5} {z}
+// CHECK:  encoding: [0x62,0x21,0x7e,0x8d,0x12,0xcd]
+          vmovsldup %xmm21, %xmm25 {%k5} {z}
+
+// CHECK: vmovsldup (%rcx), %xmm25
+// CHECK:  encoding: [0x62,0x61,0x7e,0x08,0x12,0x09]
+          vmovsldup (%rcx), %xmm25
+
+// CHECK: vmovsldup 291(%rax,%r14,8), %xmm25
+// CHECK:  encoding: [0x62,0x21,0x7e,0x08,0x12,0x8c,0xf0,0x23,0x01,0x00,0x00]
+          vmovsldup 291(%rax,%r14,8), %xmm25
+
+// CHECK: vmovsldup 2032(%rdx), %xmm25
+// CHECK:  encoding: [0x62,0x61,0x7e,0x08,0x12,0x4a,0x7f]
+          vmovsldup 2032(%rdx), %xmm25
+
+// CHECK: vmovsldup 2048(%rdx), %xmm25
+// CHECK:  encoding: [0x62,0x61,0x7e,0x08,0x12,0x8a,0x00,0x08,0x00,0x00]
+          vmovsldup 2048(%rdx), %xmm25
+
+// CHECK: vmovsldup -2048(%rdx), %xmm25
+// CHECK:  encoding: [0x62,0x61,0x7e,0x08,0x12,0x4a,0x80]
+          vmovsldup -2048(%rdx), %xmm25
+
+// CHECK: vmovsldup -2064(%rdx), %xmm25
+// CHECK:  encoding: [0x62,0x61,0x7e,0x08,0x12,0x8a,0xf0,0xf7,0xff,0xff]
+          vmovsldup -2064(%rdx), %xmm25
+
+// CHECK: vmovsldup %ymm29, %ymm24
+// CHECK:  encoding: [0x62,0x01,0x7e,0x28,0x12,0xc5]
+          vmovsldup %ymm29, %ymm24
+
+// CHECK: vmovsldup %ymm29, %ymm24 {%k5}
+// CHECK:  encoding: [0x62,0x01,0x7e,0x2d,0x12,0xc5]
+          vmovsldup %ymm29, %ymm24 {%k5}
+
+// CHECK: vmovsldup %ymm29, %ymm24 {%k5} {z}
+// CHECK:  encoding: [0x62,0x01,0x7e,0xad,0x12,0xc5]
+          vmovsldup %ymm29, %ymm24 {%k5} {z}
+
+// CHECK: vmovsldup (%rcx), %ymm24
+// CHECK:  encoding: [0x62,0x61,0x7e,0x28,0x12,0x01]
+          vmovsldup (%rcx), %ymm24
+
+// CHECK: vmovsldup 291(%rax,%r14,8), %ymm24
+// CHECK:  encoding: [0x62,0x21,0x7e,0x28,0x12,0x84,0xf0,0x23,0x01,0x00,0x00]
+          vmovsldup 291(%rax,%r14,8), %ymm24
+
+// CHECK: vmovsldup 4064(%rdx), %ymm24
+// CHECK:  encoding: [0x62,0x61,0x7e,0x28,0x12,0x42,0x7f]
+          vmovsldup 4064(%rdx), %ymm24
+
+// CHECK: vmovsldup 4096(%rdx), %ymm24
+// CHECK:  encoding: [0x62,0x61,0x7e,0x28,0x12,0x82,0x00,0x10,0x00,0x00]
+          vmovsldup 4096(%rdx), %ymm24
+
+// CHECK: vmovsldup -4096(%rdx), %ymm24
+// CHECK:  encoding: [0x62,0x61,0x7e,0x28,0x12,0x42,0x80]
+          vmovsldup -4096(%rdx), %ymm24
+
+// CHECK: vmovsldup -4128(%rdx), %ymm24
+// CHECK:  encoding: [0x62,0x61,0x7e,0x28,0x12,0x82,0xe0,0xef,0xff,0xff]
+          vmovsldup -4128(%rdx), %ymm24
+