AVX-512: intrinsics for VPADD, VPMULDQ and VPSUB

by Asaf Badouh (asaf.badouh@intel.com)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@233906 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td
index 055b87ac9c526173fe7e683d877dbc40fc59ceb9..49231d814f307894361d7d5434f39937d819e3d6 100644 (file)
--- a/include/llvm/IR/IntrinsicsX86.td
+++ b/include/llvm/IR/IntrinsicsX86.td
@@ -1408,12 +1408,6 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx2_psad_bw : GCCBuiltin<"__builtin_ia32_psadbw256">,
               Intrinsic<[llvm_v4i64_ty], [llvm_v32i8_ty,
                          llvm_v32i8_ty], [IntrNoMem, Commutative]>;
-  def int_x86_avx512_mask_pmulu_dq_512 : GCCBuiltin<"__builtin_ia32_pmuludq512_mask">,
-              Intrinsic<[llvm_v8i64_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
-                         llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
-  def int_x86_avx512_mask_pmul_dq_512 : GCCBuiltin<"__builtin_ia32_pmuldq512_mask">,
-              Intrinsic<[llvm_v8i64_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
-                         llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
 }
 
 // Vector min, max
@@ -3242,6 +3236,27 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
                       [IntrNoMem]>;
 }
 
+// Integer arithmetic ops
+let TargetPrefix = "x86" in {
+  def int_x86_avx512_mask_padd_d_512 : GCCBuiltin<"__builtin_ia32_paddd512_mask">,
+          Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
+                     llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_padd_q_512 : GCCBuiltin<"__builtin_ia32_paddq512_mask">,
+          Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
+                     llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_psub_d_512 : GCCBuiltin<"__builtin_ia32_psubd512_mask">,
+          Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
+                     llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_psub_q_512 : GCCBuiltin<"__builtin_ia32_psubq512_mask">,
+          Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
+                     llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmulu_dq_512 : GCCBuiltin<"__builtin_ia32_pmuludq512_mask">,
+              Intrinsic<[llvm_v8i64_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
+                         llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmul_dq_512 : GCCBuiltin<"__builtin_ia32_pmuldq512_mask">,
+              Intrinsic<[llvm_v8i64_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
+                         llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+}
 // Gather and Scatter ops
 let TargetPrefix = "x86" in {
   def int_x86_avx512_gather_dpd_512  : GCCBuiltin<"__builtin_ia32_gathersiv8df">,

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index 6586b6b4e13b371dad2d6b008a78eb181ab799a0..095916233c83eda1989ae4e58f4ec074e28464e9 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -2971,60 +2971,36 @@ multiclass avx512_binop_rm_vl_all<bits<8> opc_b, bits<8> opc_w,
                                     itins, HasBWI, IsCommutable>;
 }
 
-multiclass avx512_binop_rm2<bits<8> opc, string OpcodeStr, ValueType DstVT,
-                            ValueType SrcVT, RegisterClass KRC, RegisterClass RC,
-                            PatFrag memop_frag, X86MemOperand x86memop,
-                            PatFrag scalar_mfrag, X86MemOperand x86scalar_mop,
-                            string BrdcstStr, OpndItins itins, bit IsCommutable = 0> {
-  let isCommutable = IsCommutable in
-  {
-    def rr : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
-       (ins RC:$src1, RC:$src2),
-       !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-       []>, EVEX_4V;
-    def rrk : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
-               (ins KRC:$mask, RC:$src1, RC:$src2),
-               !strconcat(OpcodeStr,
-                  "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
-               [], itins.rr>, EVEX_4V, EVEX_K;
-    def rrkz : AVX512BI<opc, MRMSrcReg, (outs RC:$dst),
-                (ins KRC:$mask, RC:$src1, RC:$src2),
-                !strconcat(OpcodeStr, "\t{$src2, $src1, $dst {${mask}} {z}" ,
-                    "|$dst {${mask}} {z}, $src1, $src2}"),
-                [], itins.rr>, EVEX_4V, EVEX_KZ;
-  }
+multiclass avx512_binop_rm2<bits<8> opc, string OpcodeStr, OpndItins itins,
+                            SDNode OpNode,X86VectorVTInfo _Src, 
+                            X86VectorVTInfo _Dst, bit IsCommutable = 0> {
+  defm rr : AVX512_maskable<opc, MRMSrcReg, _Dst, (outs _Dst.RC:$dst), 
+                            (ins _Src.RC:$src1, _Src.RC:$src2), OpcodeStr,
+                            "$src2, $src1","$src1, $src2", 
+                            (_Dst.VT (OpNode 
+                                         (_Src.VT _Src.RC:$src1), 
+                                         (_Src.VT _Src.RC:$src2))),
+                            "",itins.rr, IsCommutable>, 
+                            AVX512BIBase, EVEX_4V;
   let mayLoad = 1 in {
-    def rm : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
-              (ins RC:$src1, x86memop:$src2),
-              !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
-              []>, EVEX_4V;
-    def rmk : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
-               (ins KRC:$mask, RC:$src1, x86memop:$src2),
-               !strconcat(OpcodeStr,
-                   "\t{$src2, $src1, $dst {${mask}}|$dst {${mask}}, $src1, $src2}"),
-               [], itins.rm>, EVEX_4V, EVEX_K;
-    def rmkz : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
-                (ins KRC:$mask, RC:$src1, x86memop:$src2),
-                !strconcat(OpcodeStr,
-                    "\t{$src2, $src1, $dst {${mask}} {z}|$dst {${mask}} {z}, $src1, $src2}"),
-                [], itins.rm>, EVEX_4V, EVEX_KZ;
-    def rmb : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
-               (ins RC:$src1, x86scalar_mop:$src2),
-               !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
-                          ", $src1, $dst|$dst, $src1, ${src2}", BrdcstStr, "}"),
-               [], itins.rm>, EVEX_4V, EVEX_B;
-    def rmbk : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
-                (ins KRC:$mask, RC:$src1, x86scalar_mop:$src2),
-                !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
-                           ", $src1, $dst {${mask}}|$dst {${mask}}, $src1, ${src2}",
-                           BrdcstStr, "}"),
-                [], itins.rm>, EVEX_4V, EVEX_B, EVEX_K;
-    def rmbkz : AVX512BI<opc, MRMSrcMem, (outs RC:$dst),
-                 (ins KRC:$mask, RC:$src1, x86scalar_mop:$src2),
-                 !strconcat(OpcodeStr, "\t{${src2}", BrdcstStr,
-                            ", $src1, $dst {${mask}} {z}|$dst {${mask}} {z}, $src1, ${src2}",
-                            BrdcstStr, "}"),
-                 [], itins.rm>, EVEX_4V, EVEX_B, EVEX_KZ;
+      defm rm : AVX512_maskable<opc, MRMSrcMem, _Dst, (outs _Dst.RC:$dst),
+                            (ins _Src.RC:$src1, _Src.MemOp:$src2), OpcodeStr,
+                            "$src2, $src1", "$src1, $src2",
+                            (_Dst.VT (OpNode (_Src.VT _Src.RC:$src1),
+                                          (bitconvert (_Src.LdFrag addr:$src2)))),
+                            "", itins.rm>,
+                            AVX512BIBase, EVEX_4V;
+
+      defm rmb : AVX512_maskable<opc, MRMSrcMem, _Dst, (outs _Dst.RC:$dst),
+                        (ins _Src.RC:$src1, _Dst.ScalarMemOp:$src2), 
+                        OpcodeStr,
+                        "${src2}"##_Dst.BroadcastStr##", $src1",
+                         "$src1, ${src2}"##_Dst.BroadcastStr,
+                        (_Dst.VT (OpNode (_Src.VT _Src.RC:$src1), (bc_v16i32 
+                                     (_Dst.VT (X86VBroadcast 
+                                              (_Dst.ScalarLdFrag addr:$src2)))))),
+                        "", itins.rm>,
+                        AVX512BIBase, EVEX_4V, EVEX_B;
   }
 }
 
@@ -3039,24 +3015,13 @@ defm VPMULLW : avx512_binop_rm_vl_w<0xD5, "vpmull", mul,
 defm VPMULLQ : avx512_binop_rm_vl_q<0x40, "vpmull", mul,
                                    SSE_INTALU_ITINS_P, HasDQI, 1>, T8PD;
 
-defm VPMULDQZ : avx512_binop_rm2<0x28, "vpmuldq", v8i64, v16i32, VK8WM, VR512,
-                   loadv8i64, i512mem, loadi64, i64mem, "{1to8}",
-                   SSE_INTALU_ITINS_P, 1>, T8PD, EVEX_V512,
-                   EVEX_CD8<64, CD8VF>, VEX_W;
-
-defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", v8i64, v16i32, VK8WM, VR512,
-                   loadv8i64, i512mem, loadi64, i64mem, "{1to8}",
-                   SSE_INTMUL_ITINS_P, 1>, EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
-
-def : Pat<(v8i64 (X86pmuludq (v16i32 VR512:$src1), (v16i32 VR512:$src2))),
-          (VPMULUDQZrr VR512:$src1, VR512:$src2)>;
+defm VPMULDQZ : avx512_binop_rm2<0x28, "vpmuldq", SSE_INTALU_ITINS_P,
+                   X86pmuldq, v16i32_info, v8i64_info, 1>, 
+                   T8PD, EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
 
-def : Pat<(v8i64 (int_x86_avx512_mask_pmulu_dq_512 (v16i32 VR512:$src1),
-           (v16i32 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
-          (VPMULUDQZrr VR512:$src1, VR512:$src2)>;
-def : Pat<(v8i64 (int_x86_avx512_mask_pmul_dq_512 (v16i32 VR512:$src1),
-           (v16i32 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
-          (VPMULDQZrr VR512:$src1, VR512:$src2)>;
+defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", SSE_INTMUL_ITINS_P,
+                   X86pmuludq, v16i32_info, v8i64_info, 1>, 
+                   EVEX_V512, EVEX_CD8<64, CD8VF>, VEX_W;
 
 defm VPMAXSB : avx512_binop_rm_vl_b<0x3C, "vpmaxs", X86smax,
                                      SSE_INTALU_ITINS_P, HasBWI, 1>, T8PD;

diff --git a/lib/Target/X86/X86IntrinsicsInfo.h b/lib/Target/X86/X86IntrinsicsInfo.h
index c1a51a2e19727f5bdfb129ea16be8d8f084bdd49..28a3b7b11e0a6d4b8fd645510c1785afb1c18e5e 100644 (file)
--- a/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/lib/Target/X86/X86IntrinsicsInfo.h
@@ -334,6 +334,8 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
                      X86ISD::FMUL_RND),
   X86_INTRINSIC_DATA(avx512_mask_mul_ps_512, INTR_TYPE_2OP_MASK, ISD::FMUL,
                      X86ISD::FMUL_RND),
+  X86_INTRINSIC_DATA(avx512_mask_padd_d_512, INTR_TYPE_2OP_MASK, ISD::ADD, 0),
+  X86_INTRINSIC_DATA(avx512_mask_padd_q_512, INTR_TYPE_2OP_MASK, ISD::ADD, 0),
   X86_INTRINSIC_DATA(avx512_mask_pand_d_512, INTR_TYPE_2OP_MASK, ISD::AND, 0),
   X86_INTRINSIC_DATA(avx512_mask_pand_q_512, INTR_TYPE_2OP_MASK, ISD::AND, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpeq_b_128,  CMP_MASK,  X86ISD::PCMPEQM, 0),
@@ -360,6 +362,10 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_w_128,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_w_256,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_w_512,  CMP_MASK,  X86ISD::PCMPGTM, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmul_dq_512, INTR_TYPE_2OP_MASK,
+                     X86ISD::PMULDQ, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmulu_dq_512, INTR_TYPE_2OP_MASK,
+                     X86ISD::PMULUDQ, 0),
   X86_INTRINSIC_DATA(avx512_mask_por_d_512, INTR_TYPE_2OP_MASK, ISD::OR, 0),
   X86_INTRINSIC_DATA(avx512_mask_por_q_512, INTR_TYPE_2OP_MASK, ISD::OR, 0),
   X86_INTRINSIC_DATA(avx512_mask_psll_d,        INTR_TYPE_2OP_MASK, X86ISD::VSHL, 0),
@@ -380,6 +386,8 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_psrli_q,       VSHIFT_MASK, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx512_mask_psrlv_d,       INTR_TYPE_2OP_MASK, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx512_mask_psrlv_q,       INTR_TYPE_2OP_MASK, ISD::SRL, 0),
+  X86_INTRINSIC_DATA(avx512_mask_psub_d_512, INTR_TYPE_2OP_MASK, ISD::SUB, 0),
+  X86_INTRINSIC_DATA(avx512_mask_psub_q_512, INTR_TYPE_2OP_MASK, ISD::SUB, 0),
   X86_INTRINSIC_DATA(avx512_mask_pxor_d_512, INTR_TYPE_2OP_MASK, ISD::XOR, 0),
   X86_INTRINSIC_DATA(avx512_mask_pxor_q_512, INTR_TYPE_2OP_MASK, ISD::XOR, 0),
   X86_INTRINSIC_DATA(avx512_mask_rndscale_sd,   INTR_TYPE_SCALAR_MASK_RM,

diff --git a/test/CodeGen/X86/avx512-intrinsics.ll b/test/CodeGen/X86/avx512-intrinsics.ll
index 512be32e78c7c8207a915f8187aa19b3cf31cc2c..07d984a0ee93cc3ac310918c86de64dca2a7dfd4 100644 (file)
--- a/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/test/CodeGen/X86/avx512-intrinsics.ll
@@ -515,14 +515,6 @@ define <16 x i32> @test_vpmaxsd(<16 x i32> %a0, <16 x i32> %a1) {
 }
 declare <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
 
-define <8 x i64> @test_vpmuludq(<16 x i32> %a0, <16 x i32> %a1) {
-  ; CHECK: vpmuludq {{.*}}encoding: [0x62,0xf1,0xfd,0x48,0xf4,0xc1]
-  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a0, <16 x i32> %a1,
-                    <8 x i64>zeroinitializer, i8 -1)
-  ret <8 x i64> %res
-}
-declare <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32>, <16 x i32>, <8 x i64>, i8)
-
 define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1) {
   ; CHECK: vptestmq {{.*}}encoding: [0x62,0xf2,0xfd,0x48,0x27,0xc1]
   %res = call i8 @llvm.x86.avx512.mask.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
@@ -1703,3 +1695,471 @@ define <8 x i64> @test_mask_and_epi64(<8 x i64> %a,<8 x i64> %b, <8 x i64> %pass
 
 declare <8 x i64> @llvm.x86.avx512.mask.pand.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
 
+
+define <16 x i32> @test_mask_add_epi32_rr(<16 x i32> %a, <16 x i32> %b) {
+  ;CHECK-LABEL: test_mask_add_epi32_rr
+  ;CHECK: vpaddd %zmm1, %zmm0, %zmm0     ## encoding: [0x62,0xf1,0x7d,0x48,0xfe,0xc1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 -1)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rrk(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi32_rrk
+  ;CHECK: vpaddd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x49,0xfe,0xd1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rrkz(<16 x i32> %a, <16 x i32> %b, i16 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi32_rrkz
+  ;CHECK: vpaddd %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xc9,0xfe,0xc1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rm(<16 x i32> %a, <16 x i32>* %ptr_b) {
+  ;CHECK-LABEL: test_mask_add_epi32_rm
+  ;CHECK: vpaddd (%rdi), %zmm0, %zmm0    ## encoding: [0x62,0xf1,0x7d,0x48,0xfe,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 -1)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rmk(<16 x i32> %a, <16 x i32>* %ptr_b, <16 x i32> %passThru, i16 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi32_rmk
+  ;CHECK: vpaddd (%rdi), %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x49,0xfe,0x0f]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rmkz(<16 x i32> %a, <16 x i32>* %ptr_b, i16 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi32_rmkz
+  ;CHECK: vpaddd (%rdi), %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xc9,0xfe,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rmb(<16 x i32> %a, i32* %ptr_b) {
+  ;CHECK-LABEL: test_mask_add_epi32_rmb
+  ;CHECK: vpaddd (%rdi){1to16}, %zmm0, %zmm0  ## encoding: [0x62,0xf1,0x7d,0x58,0xfe,0x07]
+  %q = load i32, i32* %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 -1)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rmbk(<16 x i32> %a, i32* %ptr_b, <16 x i32> %passThru, i16 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi32_rmbk
+  ;CHECK: vpaddd (%rdi){1to16}, %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x59,0xfe,0x0f]
+  %q = load i32, i32* %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_add_epi32_rmbkz(<16 x i32> %a, i32* %ptr_b, i16 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi32_rmbkz
+  ;CHECK: vpaddd (%rdi){1to16}, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xd9,0xfe,0x07]
+  %q = load i32, i32* %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx512.mask.padd.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+define <16 x i32> @test_mask_sub_epi32_rr(<16 x i32> %a, <16 x i32> %b) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rr
+  ;CHECK: vpsubd %zmm1, %zmm0, %zmm0     ## encoding: [0x62,0xf1,0x7d,0x48,0xfa,0xc1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 -1)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rrk(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rrk
+  ;CHECK: vpsubd %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x49,0xfa,0xd1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rrkz(<16 x i32> %a, <16 x i32> %b, i16 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rrkz
+  ;CHECK: vpsubd %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xc9,0xfa,0xc1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rm(<16 x i32> %a, <16 x i32>* %ptr_b) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rm
+  ;CHECK: vpsubd (%rdi), %zmm0, %zmm0    ## encoding: [0x62,0xf1,0x7d,0x48,0xfa,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 -1)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rmk(<16 x i32> %a, <16 x i32>* %ptr_b, <16 x i32> %passThru, i16 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rmk
+  ;CHECK: vpsubd (%rdi), %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x49,0xfa,0x0f]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rmkz(<16 x i32> %a, <16 x i32>* %ptr_b, i16 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rmkz
+  ;CHECK: vpsubd (%rdi), %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xc9,0xfa,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rmb(<16 x i32> %a, i32* %ptr_b) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rmb
+  ;CHECK: vpsubd (%rdi){1to16}, %zmm0, %zmm0  ## encoding: [0x62,0xf1,0x7d,0x58,0xfa,0x07]
+  %q = load i32, i32* %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 -1)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rmbk(<16 x i32> %a, i32* %ptr_b, <16 x i32> %passThru, i16 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rmbk
+  ;CHECK: vpsubd (%rdi){1to16}, %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x59,0xfa,0x0f]
+  %q = load i32, i32* %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> %passThru, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+define <16 x i32> @test_mask_sub_epi32_rmbkz(<16 x i32> %a, i32* %ptr_b, i16 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi32_rmbkz
+  ;CHECK: vpsubd (%rdi){1to16}, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xd9,0xfa,0x07]
+  %q = load i32, i32* %ptr_b
+  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
+  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
+  %res = call <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32> %a, <16 x i32> %b, <16 x i32> zeroinitializer, i16 %mask)
+  ret < 16 x i32> %res
+}
+
+declare <16 x i32> @llvm.x86.avx512.mask.psub.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+define <8 x i64> @test_mask_add_epi64_rr(<8 x i64> %a, <8 x i64> %b) {
+  ;CHECK-LABEL: test_mask_add_epi64_rr
+  ;CHECK: vpaddq %zmm1, %zmm0, %zmm0     ## encoding: [0x62,0xf1,0xfd,0x48,0xd4,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rrk(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi64_rrk
+  ;CHECK: vpaddq %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf1,0xfd,0x49,0xd4,0xd1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rrkz(<8 x i64> %a, <8 x i64> %b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi64_rrkz
+  ;CHECK: vpaddq %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xd4,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rm(<8 x i64> %a, <8 x i64>* %ptr_b) {
+  ;CHECK-LABEL: test_mask_add_epi64_rm
+  ;CHECK: vpaddq (%rdi), %zmm0, %zmm0    ## encoding: [0x62,0xf1,0xfd,0x48,0xd4,0x07]
+  %b = load <8 x i64>, <8 x i64>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rmk(<8 x i64> %a, <8 x i64>* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi64_rmk
+  ;CHECK: vpaddq (%rdi), %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x49,0xd4,0x0f]
+  %b = load <8 x i64>, <8 x i64>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rmkz(<8 x i64> %a, <8 x i64>* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi64_rmkz
+  ;CHECK: vpaddq (%rdi), %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xd4,0x07]
+  %b = load <8 x i64>, <8 x i64>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rmb(<8 x i64> %a, i64* %ptr_b) {
+  ;CHECK-LABEL: test_mask_add_epi64_rmb
+  ;CHECK: vpaddq (%rdi){1to8}, %zmm0, %zmm0  ## encoding: [0x62,0xf1,0xfd,0x58,0xd4,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rmbk(<8 x i64> %a, i64* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi64_rmbk
+  ;CHECK: vpaddq (%rdi){1to8}, %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xd4,0x0f]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_add_epi64_rmbkz(<8 x i64> %a, i64* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_add_epi64_rmbkz
+  ;CHECK: vpaddq (%rdi){1to8}, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xd9,0xd4,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %res = call <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.padd.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
+
+define <8 x i64> @test_mask_sub_epi64_rr(<8 x i64> %a, <8 x i64> %b) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rr
+  ;CHECK: vpsubq %zmm1, %zmm0, %zmm0     ## encoding: [0x62,0xf1,0xfd,0x48,0xfb,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rrk(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rrk
+  ;CHECK: vpsubq %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf1,0xfd,0x49,0xfb,0xd1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rrkz(<8 x i64> %a, <8 x i64> %b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rrkz
+  ;CHECK: vpsubq %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xfb,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rm(<8 x i64> %a, <8 x i64>* %ptr_b) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rm
+  ;CHECK: vpsubq (%rdi), %zmm0, %zmm0    ## encoding: [0x62,0xf1,0xfd,0x48,0xfb,0x07]
+  %b = load <8 x i64>, <8 x i64>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rmk(<8 x i64> %a, <8 x i64>* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rmk
+  ;CHECK: vpsubq (%rdi), %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x49,0xfb,0x0f]
+  %b = load <8 x i64>, <8 x i64>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rmkz(<8 x i64> %a, <8 x i64>* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rmkz
+  ;CHECK: vpsubq (%rdi), %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xfb,0x07]
+  %b = load <8 x i64>, <8 x i64>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rmb(<8 x i64> %a, i64* %ptr_b) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rmb
+  ;CHECK: vpsubq (%rdi){1to8}, %zmm0, %zmm0  ## encoding: [0x62,0xf1,0xfd,0x58,0xfb,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rmbk(<8 x i64> %a, i64* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rmbk
+  ;CHECK: vpsubq (%rdi){1to8}, %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xfb,0x0f]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_sub_epi64_rmbkz(<8 x i64> %a, i64* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_sub_epi64_rmbkz
+  ;CHECK: vpsubq (%rdi){1to8}, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xd9,0xfb,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %res = call <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64> %a, <8 x i64> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.psub.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
+
+define <8 x i64> @test_mask_mul_epi32_rr(<16 x i32> %a, <16 x i32> %b) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rr
+  ;CHECK: vpmuldq %zmm1, %zmm0, %zmm0     ## encoding: [0x62,0xf2,0xfd,0x48,0x28,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rrk(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rrk
+  ;CHECK: vpmuldq %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0x28,0xd1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rrkz(<16 x i32> %a, <16 x i32> %b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rrkz
+  ;CHECK: vpmuldq %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xc9,0x28,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rm(<16 x i32> %a, <16 x i32>* %ptr_b) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rm
+  ;CHECK: vpmuldq (%rdi), %zmm0, %zmm0    ## encoding: [0x62,0xf2,0xfd,0x48,0x28,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rmk(<16 x i32> %a, <16 x i32>* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rmk
+  ;CHECK: vpmuldq (%rdi), %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x49,0x28,0x0f]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rmkz(<16 x i32> %a, <16 x i32>* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rmkz
+  ;CHECK: vpmuldq (%rdi), %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xc9,0x28,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rmb(<16 x i32> %a, i64* %ptr_b) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rmb
+  ;CHECK: vpmuldq (%rdi){1to8}, %zmm0, %zmm0  ## encoding: [0x62,0xf2,0xfd,0x58,0x28,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %b = bitcast <8 x i64> %b64 to <16 x i32>
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rmbk(<16 x i32> %a, i64* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rmbk
+  ;CHECK: vpmuldq (%rdi){1to8}, %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf2,0xfd,0x59,0x28,0x0f]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %b = bitcast <8 x i64> %b64 to <16 x i32>
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epi32_rmbkz(<16 x i32> %a, i64* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epi32_rmbkz
+  ;CHECK: vpmuldq (%rdi){1to8}, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xd9,0x28,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %b = bitcast <8 x i64> %b64 to <16 x i32>
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.pmul.dq.512(<16 x i32>, <16 x i32>, <8 x i64>, i8)
+
+define <8 x i64> @test_mask_mul_epu32_rr(<16 x i32> %a, <16 x i32> %b) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rr
+  ;CHECK: vpmuludq %zmm1, %zmm0, %zmm0  ## encoding: [0x62,0xf1,0xfd,0x48,0xf4,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rrk(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rrk
+  ;CHECK: vpmuludq %zmm1, %zmm0, %zmm2 {%k1} ## encoding: [0x62,0xf1,0xfd,0x49,0xf4,0xd1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rrkz(<16 x i32> %a, <16 x i32> %b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rrkz
+  ;CHECK: vpmuludq %zmm1, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xf4,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rm(<16 x i32> %a, <16 x i32>* %ptr_b) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rm
+  ;CHECK: vpmuludq (%rdi), %zmm0, %zmm0  ## encoding: [0x62,0xf1,0xfd,0x48,0xf4,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rmk(<16 x i32> %a, <16 x i32>* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rmk
+  ;CHECK: vpmuludq (%rdi), %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x49,0xf4,0x0f]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rmkz(<16 x i32> %a, <16 x i32>* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rmkz
+  ;CHECK: vpmuludq (%rdi), %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xc9,0xf4,0x07]
+  %b = load <16 x i32>, <16 x i32>* %ptr_b
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rmb(<16 x i32> %a, i64* %ptr_b) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rmb
+  ;CHECK: vpmuludq (%rdi){1to8}, %zmm0, %zmm0  ## encoding: [0x62,0xf1,0xfd,0x58,0xf4,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %b = bitcast <8 x i64> %b64 to <16 x i32>
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 -1)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rmbk(<16 x i32> %a, i64* %ptr_b, <8 x i64> %passThru, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rmbk
+  ;CHECK: vpmuludq (%rdi){1to8}, %zmm0, %zmm1 {%k1} ## encoding: [0x62,0xf1,0xfd,0x59,0xf4,0x0f]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %b = bitcast <8 x i64> %b64 to <16 x i32>
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+define <8 x i64> @test_mask_mul_epu32_rmbkz(<16 x i32> %a, i64* %ptr_b, i8 %mask) {
+  ;CHECK-LABEL: test_mask_mul_epu32_rmbkz
+  ;CHECK: vpmuludq (%rdi){1to8}, %zmm0, %zmm0 {%k1} {z} ## encoding: [0x62,0xf1,0xfd,0xd9,0xf4,0x07]
+  %q = load i64, i64* %ptr_b
+  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
+  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
+  %b = bitcast <8 x i64> %b64 to <16 x i32>
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b, <8 x i64> zeroinitializer, i8 %mask)
+  ret < 8 x i64> %res
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.pmulu.dq.512(<16 x i32>, <16 x i32>, <8 x i64>, i8)