AVX512: Implemented encoding and intrinsics for vdbpsadbw

Added tests for intrinsics and encoding.

Differential Revision: http://reviews.llvm.org/D12491

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@246436 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td
index e78dc75db713eb4ccdf2dfcecbcb376c34f01177..bfd1279b1d99eeb65618241eb94ac85737d7a774 100644 (file)
--- a/include/llvm/IR/IntrinsicsX86.td
+++ b/include/llvm/IR/IntrinsicsX86.td
@@ -5232,6 +5232,24 @@ let TargetPrefix = "x86" in {
           Intrinsic<[llvm_v32i16_ty],
           [llvm_v64i8_ty, llvm_v64i8_ty, llvm_v32i16_ty,  llvm_i32_ty],
           [IntrNoMem]>;
+
+  def int_x86_avx512_mask_dbpsadbw_128 :
+         GCCBuiltin<"__builtin_ia32_dbpsadbw128_mask">,
+          Intrinsic<[llvm_v8i16_ty],
+          [llvm_v16i8_ty, llvm_v16i8_ty, llvm_i32_ty, llvm_v8i16_ty,
+           llvm_i8_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_mask_dbpsadbw_256 :
+         GCCBuiltin<"__builtin_ia32_dbpsadbw256_mask">,
+          Intrinsic<[llvm_v16i16_ty],
+          [llvm_v32i8_ty, llvm_v32i8_ty, llvm_i32_ty, llvm_v16i16_ty,
+           llvm_i16_ty], [IntrNoMem]>;
+
+  def int_x86_avx512_mask_dbpsadbw_512 :
+         GCCBuiltin<"__builtin_ia32_dbpsadbw512_mask">,
+          Intrinsic<[llvm_v32i16_ty],
+          [llvm_v64i8_ty, llvm_v64i8_ty, llvm_i32_ty, llvm_v32i16_ty,
+           llvm_i32_ty], [IntrNoMem]>;
 }
 
 // Gather and Scatter ops

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 1f591e0b51d2aa9a9f19b8464b8d31c17c5e0229..9817e23ef119e6fc9a2fa59c0eb78b35c75369f4 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -19353,6 +19353,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::PMULUDQ:            return "X86ISD::PMULUDQ";
   case X86ISD::PMULDQ:             return "X86ISD::PMULDQ";
   case X86ISD::PSADBW:             return "X86ISD::PSADBW";
+  case X86ISD::DBPSADBW:           return "X86ISD::DBPSADBW";
   case X86ISD::VASTART_SAVE_XMM_REGS: return "X86ISD::VASTART_SAVE_XMM_REGS";
   case X86ISD::VAARG_64:           return "X86ISD::VAARG_64";
   case X86ISD::WIN_ALLOCA:         return "X86ISD::WIN_ALLOCA";

diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 69bb0e3b0e4165cf907b4cc65166f892ef266194..daa1ab23677f944e4619461b2c765dfaa465c8bc 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -182,6 +182,8 @@ namespace llvm {
 
       /// Compute Sum of Absolute Differences.
       PSADBW,
+      /// Compute Double Block Packed Sum-Absolute-Differences
+      DBPSADBW,
 
       /// Bitwise Logical AND NOT of Packed FP values.
       ANDNP,

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index b2ac33781f403416febed4d93dfda8829dca5e6c..3c11e96f8644ed3078b9e29b42be5ee134285d32 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -6713,6 +6713,9 @@ defm VPALIGN:   avx512_common_3Op_rm_imm8<0x0F, X86PAlignr, "vpalignr" ,
                 avx512_vpalign_lowering_common<avx512vl_f64_info>,
                 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),

diff --git a/lib/Target/X86/X86InstrFragmentsSIMD.td b/lib/Target/X86/X86InstrFragmentsSIMD.td
index 361e1966ae717bf0c863b0e3913e30d2e50d174d..5cd04d8434780147bb2bd4ac545c2c837a7a616d 100644 (file)
--- a/lib/Target/X86/X86InstrFragmentsSIMD.td
+++ b/lib/Target/X86/X86InstrFragmentsSIMD.td
@@ -79,6 +79,9 @@ def X86pshufb  : SDNode<"X86ISD::PSHUFB",
 def X86psadbw  : SDNode<"X86ISD::PSADBW",
                  SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                       SDTCisSameAs<0,2>]>>;
+def X86dbpsadbw : SDNode<"X86ISD::DBPSADBW",
+                  SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisVec<1>,
+                                       SDTCisSameAs<1,2>, SDTCisInt<3>]>>;
 def X86andnp   : SDNode<"X86ISD::ANDNP",
                  SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0,1>,
                                       SDTCisSameAs<0,2>]>>;

diff --git a/lib/Target/X86/X86IntrinsicsInfo.h b/lib/Target/X86/X86IntrinsicsInfo.h
index ed5162941397c82088e0206da24936a39e5b30b4..e2958429cbced4d9086d4e5db3b7a24615827f39 100644 (file)
--- a/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/lib/Target/X86/X86IntrinsicsInfo.h
@@ -611,7 +611,12 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
                      ISD::UINT_TO_FP, 0),
   X86_INTRINSIC_DATA(avx512_mask_cvtuqq2ps_512, INTR_TYPE_1OP_MASK,
                      ISD::UINT_TO_FP, ISD::UINT_TO_FP),
-
+  X86_INTRINSIC_DATA(avx512_mask_dbpsadbw_128, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::DBPSADBW, 0),
+  X86_INTRINSIC_DATA(avx512_mask_dbpsadbw_256, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::DBPSADBW, 0),
+  X86_INTRINSIC_DATA(avx512_mask_dbpsadbw_512, INTR_TYPE_3OP_IMM8_MASK,
+                     X86ISD::DBPSADBW, 0),
   X86_INTRINSIC_DATA(avx512_mask_div_pd_128, INTR_TYPE_2OP_MASK, ISD::FDIV, 0),
   X86_INTRINSIC_DATA(avx512_mask_div_pd_256, INTR_TYPE_2OP_MASK, ISD::FDIV, 0),
   X86_INTRINSIC_DATA(avx512_mask_div_pd_512, INTR_TYPE_2OP_MASK, ISD::FDIV,

diff --git a/test/CodeGen/X86/avx512bw-intrinsics.ll b/test/CodeGen/X86/avx512bw-intrinsics.ll
index 981fbfcb5cf5da22443458e6e1098e09271028ad..ab724da69272ee377a9a207f705cef8bca8c126e 100644 (file)
--- a/test/CodeGen/X86/avx512bw-intrinsics.ll
+++ b/test/CodeGen/X86/avx512bw-intrinsics.ll
@@ -1201,3 +1201,23 @@ define <64 x i8>@test_int_x86_avx512_mask_palignr_512(<64 x i8> %x0, <64 x i8> %
   %res4 = add <64 x i8> %res3, %res2
   ret <64 x i8> %res4
 }
+
+declare <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8>, <64 x i8>, i32, <32 x i16>, i32)
+
+define <32 x i16>@test_int_x86_avx512_mask_dbpsadbw_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x3, i32 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_dbpsadbw_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovd %edi, %k1
+; CHECK-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm3 {%k1} {z}
+; CHECK-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    vpaddw %zmm3, %zmm2, %zmm1
+; CHECK-NEXT:    vpaddw %zmm0, %zmm1, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <32 x i16> %x3, i32 %x4)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <32 x i16> zeroinitializer, i32 %x4)
+  %res2 = call <32 x i16> @llvm.x86.avx512.mask.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2, <32 x i16> %x3, i32 -1)
+  %res3 = add <32 x i16> %res, %res1
+  %res4 = add <32 x i16> %res3, %res2
+  ret <32 x i16> %res4
+}

diff --git a/test/CodeGen/X86/avx512bwvl-intrinsics.ll b/test/CodeGen/X86/avx512bwvl-intrinsics.ll
index d2a7f97f7a6c0c9a1c7455edc680ee818b20e492..3397c770d42075f319f960a53c23fd38d142627e 100644 (file)
--- a/test/CodeGen/X86/avx512bwvl-intrinsics.ll
+++ b/test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -4234,3 +4234,44 @@ define <32 x i8>@test_int_x86_avx512_mask_palignr_256(<32 x i8> %x0, <32 x i8> %
   %res4 = add <32 x i8> %res3, %res2
   ret <32 x i8> %res4
 }
+
+declare <8 x i16> @llvm.x86.avx512.mask.dbpsadbw.128(<16 x i8>, <16 x i8>, i32, <8 x i16>, i8)
+
+define <8 x i16>@test_int_x86_avx512_mask_dbpsadbw_128(<16 x i8> %x0, <16 x i8> %x1, <8 x i16> %x3, i8 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_dbpsadbw_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    kmovw %eax, %k1
+; CHECK-NEXT:    vdbpsadbw $2, %xmm1, %xmm0, %xmm2 {%k1}
+; CHECK-NEXT:    vdbpsadbw $2, %xmm1, %xmm0, %xmm3 {%k1} {z}
+; CHECK-NEXT:    vdbpsadbw $2, %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vpaddw %xmm3, %xmm2, %xmm1
+; CHECK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i16> @llvm.x86.avx512.mask.dbpsadbw.128(<16 x i8> %x0, <16 x i8> %x1, i32 2, <8 x i16> %x3, i8 %x4)
+  %res1 = call <8 x i16> @llvm.x86.avx512.mask.dbpsadbw.128(<16 x i8> %x0, <16 x i8> %x1, i32 2, <8 x i16> zeroinitializer, i8 %x4)
+  %res2 = call <8 x i16> @llvm.x86.avx512.mask.dbpsadbw.128(<16 x i8> %x0, <16 x i8> %x1, i32 2, <8 x i16> %x3, i8 -1)
+  %res3 = add <8 x i16> %res, %res1
+  %res4 = add <8 x i16> %res2, %res3
+  ret <8 x i16> %res4
+}
+
+declare <16 x i16> @llvm.x86.avx512.mask.dbpsadbw.256(<32 x i8>, <32 x i8>, i32, <16 x i16>, i16)
+
+define <16 x i16>@test_int_x86_avx512_mask_dbpsadbw_256(<32 x i8> %x0, <32 x i8> %x1, <16 x i16> %x3, i16 %x4) {
+; CHECK-LABEL: test_int_x86_avx512_mask_dbpsadbw_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vdbpsadbw $2, %ymm1, %ymm0, %ymm2 {%k1}
+; CHECK-NEXT:    vdbpsadbw $2, %ymm1, %ymm0, %ymm3 {%k1} {z}
+; CHECK-NEXT:    vdbpsadbw $2, %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpaddw %ymm3, %ymm2, %ymm1
+; CHECK-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i16> @llvm.x86.avx512.mask.dbpsadbw.256(<32 x i8> %x0, <32 x i8> %x1, i32 2, <16 x i16> %x3, i16 %x4)
+  %res1 = call <16 x i16> @llvm.x86.avx512.mask.dbpsadbw.256(<32 x i8> %x0, <32 x i8> %x1, i32 2, <16 x i16> zeroinitializer, i16 %x4)
+  %res2 = call <16 x i16> @llvm.x86.avx512.mask.dbpsadbw.256(<32 x i8> %x0, <32 x i8> %x1, i32 2, <16 x i16> %x3, i16 -1)
+  %res3 = add <16 x i16> %res, %res1
+  %res4 = add <16 x i16> %res3, %res2
+  ret <16 x i16> %res4
+}

diff --git a/test/MC/X86/x86-64-avx512bw.s b/test/MC/X86/x86-64-avx512bw.s
index acc4eafb220b24a30440243749ae1d4d7c54ecee..c83e525e2829e3272940649da9a4f570efcd9b05 100644 (file)
--- a/test/MC/X86/x86-64-avx512bw.s
+++ b/test/MC/X86/x86-64-avx512bw.s
@@ -4160,3 +4160,43 @@
 // CHECK:  encoding: [0x62,0xe3,0x2d,0x40,0x0f,0xb2,0xc0,0xdf,0xff,0xff,0x7b]
           vpalignr $123, -8256(%rdx), %zmm26, %zmm22
 
+// CHECK: vdbpsadbw $171, %zmm18, %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xa3,0x5d,0x40,0x42,0xea,0xab]
+          vdbpsadbw $171, %zmm18, %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $171, %zmm18, %zmm20, %zmm21 {%k1}
+// CHECK:  encoding: [0x62,0xa3,0x5d,0x41,0x42,0xea,0xab]
+          vdbpsadbw $171, %zmm18, %zmm20, %zmm21 {%k1}
+
+// CHECK: vdbpsadbw $171, %zmm18, %zmm20, %zmm21 {%k1} {z}
+// CHECK:  encoding: [0x62,0xa3,0x5d,0xc1,0x42,0xea,0xab]
+          vdbpsadbw $171, %zmm18, %zmm20, %zmm21 {%k1} {z}
+
+// CHECK: vdbpsadbw $123, %zmm18, %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xa3,0x5d,0x40,0x42,0xea,0x7b]
+          vdbpsadbw $123, %zmm18, %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $123, (%rcx), %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xe3,0x5d,0x40,0x42,0x29,0x7b]
+          vdbpsadbw $123, (%rcx), %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $123, 291(%rax,%r14,8), %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xa3,0x5d,0x40,0x42,0xac,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vdbpsadbw $123, 291(%rax,%r14,8), %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $123, 8128(%rdx), %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xe3,0x5d,0x40,0x42,0x6a,0x7f,0x7b]
+          vdbpsadbw $123, 8128(%rdx), %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $123, 8192(%rdx), %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xe3,0x5d,0x40,0x42,0xaa,0x00,0x20,0x00,0x00,0x7b]
+          vdbpsadbw $123, 8192(%rdx), %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $123, -8192(%rdx), %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xe3,0x5d,0x40,0x42,0x6a,0x80,0x7b]
+          vdbpsadbw $123, -8192(%rdx), %zmm20, %zmm21
+
+// CHECK: vdbpsadbw $123, -8256(%rdx), %zmm20, %zmm21
+// CHECK:  encoding: [0x62,0xe3,0x5d,0x40,0x42,0xaa,0xc0,0xdf,0xff,0xff,0x7b]
+          vdbpsadbw $123, -8256(%rdx), %zmm20, %zmm21
+

diff --git a/test/MC/X86/x86-64-avx512bw_vl.s b/test/MC/X86/x86-64-avx512bw_vl.s
index aa0f3e8583493cb9d22d936f3d438b71b0256e41..ae46e9424764dc7b6d549a0e6329272d068a23f4 100644 (file)
--- a/test/MC/X86/x86-64-avx512bw_vl.s
+++ b/test/MC/X86/x86-64-avx512bw_vl.s
@@ -8559,3 +8559,162 @@
 // CHECK:  encoding: [0x62,0xe3,0x75,0x20,0x0f,0xaa,0xe0,0xef,0xff,0xff,0x7b]
           vpalignr $0x7b,-4128(%rdx), %ymm17, %ymm21
 
+// CHECK: vdbpsadbw $171, %xmm20, %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xa3,0x15,0x00,0x42,0xcc,0xab]
+          vdbpsadbw $0xab, %xmm20, %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $171, %xmm20, %xmm29, %xmm17 {%k4}
+// CHECK:  encoding: [0x62,0xa3,0x15,0x04,0x42,0xcc,0xab]
+          vdbpsadbw $0xab, %xmm20, %xmm29, %xmm17 {%k4}
+
+// CHECK: vdbpsadbw $171, %xmm20, %xmm29, %xmm17 {%k4} {z}
+// CHECK:  encoding: [0x62,0xa3,0x15,0x84,0x42,0xcc,0xab]
+          vdbpsadbw $0xab, %xmm20, %xmm29, %xmm17 {%k4} {z}
+
+// CHECK: vdbpsadbw $123, %xmm20, %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xa3,0x15,0x00,0x42,0xcc,0x7b]
+          vdbpsadbw $0x7b, %xmm20, %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $123, (%rcx), %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xe3,0x15,0x00,0x42,0x09,0x7b]
+          vdbpsadbw $0x7b,(%rcx), %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $123, 4660(%rax,%r14,8), %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xa3,0x15,0x00,0x42,0x8c,0xf0,0x34,0x12,0x00,0x00,0x7b]
+          vdbpsadbw $0x7b,4660(%rax,%r14,8), %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $123, 2032(%rdx), %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xe3,0x15,0x00,0x42,0x4a,0x7f,0x7b]
+          vdbpsadbw $0x7b,2032(%rdx), %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $123, 2048(%rdx), %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xe3,0x15,0x00,0x42,0x8a,0x00,0x08,0x00,0x00,0x7b]
+          vdbpsadbw $0x7b,2048(%rdx), %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $123, -2048(%rdx), %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xe3,0x15,0x00,0x42,0x4a,0x80,0x7b]
+          vdbpsadbw $0x7b,-2048(%rdx), %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $123, -2064(%rdx), %xmm29, %xmm17
+// CHECK:  encoding: [0x62,0xe3,0x15,0x00,0x42,0x8a,0xf0,0xf7,0xff,0xff,0x7b]
+          vdbpsadbw $0x7b,-2064(%rdx), %xmm29, %xmm17
+
+// CHECK: vdbpsadbw $171, %ymm26, %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x03,0x1d,0x20,0x42,0xd2,0xab]
+          vdbpsadbw $0xab, %ymm26, %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $171, %ymm26, %ymm28, %ymm26 {%k4}
+// CHECK:  encoding: [0x62,0x03,0x1d,0x24,0x42,0xd2,0xab]
+          vdbpsadbw $0xab, %ymm26, %ymm28, %ymm26 {%k4}
+
+// CHECK: vdbpsadbw $171, %ymm26, %ymm28, %ymm26 {%k4} {z}
+// CHECK:  encoding: [0x62,0x03,0x1d,0xa4,0x42,0xd2,0xab]
+          vdbpsadbw $0xab, %ymm26, %ymm28, %ymm26 {%k4} {z}
+
+// CHECK: vdbpsadbw $123, %ymm26, %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x03,0x1d,0x20,0x42,0xd2,0x7b]
+          vdbpsadbw $0x7b, %ymm26, %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $123, (%rcx), %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x63,0x1d,0x20,0x42,0x11,0x7b]
+          vdbpsadbw $0x7b,(%rcx), %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $123, 4660(%rax,%r14,8), %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x23,0x1d,0x20,0x42,0x94,0xf0,0x34,0x12,0x00,0x00,0x7b]
+          vdbpsadbw $0x7b,4660(%rax,%r14,8), %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $123, 4064(%rdx), %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x63,0x1d,0x20,0x42,0x52,0x7f,0x7b]
+          vdbpsadbw $0x7b,4064(%rdx), %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $123, 4096(%rdx), %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x63,0x1d,0x20,0x42,0x92,0x00,0x10,0x00,0x00,0x7b]
+          vdbpsadbw $0x7b,4096(%rdx), %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $123, -4096(%rdx), %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x63,0x1d,0x20,0x42,0x52,0x80,0x7b]
+          vdbpsadbw $0x7b,-4096(%rdx), %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $123, -4128(%rdx), %ymm28, %ymm26
+// CHECK:  encoding: [0x62,0x63,0x1d,0x20,0x42,0x92,0xe0,0xef,0xff,0xff,0x7b]
+          vdbpsadbw $0x7b,-4128(%rdx), %ymm28, %ymm26
+
+// CHECK: vdbpsadbw $171, %xmm17, %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xa3,0x35,0x00,0x42,0xf1,0xab]
+          vdbpsadbw $171, %xmm17, %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $171, %xmm17, %xmm25, %xmm22 {%k3}
+// CHECK:  encoding: [0x62,0xa3,0x35,0x03,0x42,0xf1,0xab]
+          vdbpsadbw $171, %xmm17, %xmm25, %xmm22 {%k3}
+
+// CHECK: vdbpsadbw $171, %xmm17, %xmm25, %xmm22 {%k3} {z}
+// CHECK:  encoding: [0x62,0xa3,0x35,0x83,0x42,0xf1,0xab]
+          vdbpsadbw $171, %xmm17, %xmm25, %xmm22 {%k3} {z}
+
+// CHECK: vdbpsadbw $123, %xmm17, %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xa3,0x35,0x00,0x42,0xf1,0x7b]
+          vdbpsadbw $123, %xmm17, %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $123, (%rcx), %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xe3,0x35,0x00,0x42,0x31,0x7b]
+          vdbpsadbw $123, (%rcx), %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $123, 291(%rax,%r14,8), %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xa3,0x35,0x00,0x42,0xb4,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vdbpsadbw $123, 291(%rax,%r14,8), %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $123, 2032(%rdx), %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xe3,0x35,0x00,0x42,0x72,0x7f,0x7b]
+          vdbpsadbw $123, 2032(%rdx), %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $123, 2048(%rdx), %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xe3,0x35,0x00,0x42,0xb2,0x00,0x08,0x00,0x00,0x7b]
+          vdbpsadbw $123, 2048(%rdx), %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $123, -2048(%rdx), %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xe3,0x35,0x00,0x42,0x72,0x80,0x7b]
+          vdbpsadbw $123, -2048(%rdx), %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $123, -2064(%rdx), %xmm25, %xmm22
+// CHECK:  encoding: [0x62,0xe3,0x35,0x00,0x42,0xb2,0xf0,0xf7,0xff,0xff,0x7b]
+          vdbpsadbw $123, -2064(%rdx), %xmm25, %xmm22
+
+// CHECK: vdbpsadbw $171, %ymm20, %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xa3,0x65,0x20,0x42,0xcc,0xab]
+          vdbpsadbw $171, %ymm20, %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $171, %ymm20, %ymm19, %ymm17 {%k5}
+// CHECK:  encoding: [0x62,0xa3,0x65,0x25,0x42,0xcc,0xab]
+          vdbpsadbw $171, %ymm20, %ymm19, %ymm17 {%k5}
+
+// CHECK: vdbpsadbw $171, %ymm20, %ymm19, %ymm17 {%k5} {z}
+// CHECK:  encoding: [0x62,0xa3,0x65,0xa5,0x42,0xcc,0xab]
+          vdbpsadbw $171, %ymm20, %ymm19, %ymm17 {%k5} {z}
+
+// CHECK: vdbpsadbw $123, %ymm20, %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xa3,0x65,0x20,0x42,0xcc,0x7b]
+          vdbpsadbw $123, %ymm20, %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $123, (%rcx), %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xe3,0x65,0x20,0x42,0x09,0x7b]
+          vdbpsadbw $123, (%rcx), %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $123, 291(%rax,%r14,8), %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xa3,0x65,0x20,0x42,0x8c,0xf0,0x23,0x01,0x00,0x00,0x7b]
+          vdbpsadbw $123, 291(%rax,%r14,8), %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $123, 4064(%rdx), %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xe3,0x65,0x20,0x42,0x4a,0x7f,0x7b]
+          vdbpsadbw $123, 4064(%rdx), %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $123, 4096(%rdx), %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xe3,0x65,0x20,0x42,0x8a,0x00,0x10,0x00,0x00,0x7b]
+          vdbpsadbw $123, 4096(%rdx), %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $123, -4096(%rdx), %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xe3,0x65,0x20,0x42,0x4a,0x80,0x7b]
+          vdbpsadbw $123, -4096(%rdx), %ymm19, %ymm17
+
+// CHECK: vdbpsadbw $123, -4128(%rdx), %ymm19, %ymm17
+// CHECK:  encoding: [0x62,0xe3,0x65,0x20,0x42,0x8a,0xe0,0xef,0xff,0xff,0x7b]
+          vdbpsadbw $123, -4128(%rdx), %ymm19, %ymm17