Index: include/llvm/IR/IntrinsicsX86.td
===================================================================
--- include/llvm/IR/IntrinsicsX86.td
+++ include/llvm/IR/IntrinsicsX86.td
@@ -4019,6 +4019,12 @@
   def int_x86_avx512_kunpck_bw : GCCBuiltin<"__builtin_ia32_kunpckhi">,
               Intrinsic<[llvm_i16_ty], [llvm_i16_ty, llvm_i16_ty],
                          [IntrNoMem]>;
+  def int_x86_avx512_kunpck_wd : GCCBuiltin<"__builtin_ia32_kunpcksi">,
+              Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
+                         [IntrNoMem]>;
+  def int_x86_avx512_kunpck_dq : GCCBuiltin<"__builtin_ia32_kunpckdi">,
+              Intrinsic<[llvm_i64_ty], [llvm_i64_ty, llvm_i64_ty],
+                         [IntrNoMem]>;
   def int_x86_avx512_kortestz_w : GCCBuiltin<"__builtin_ia32_kortestzhi">,
               Intrinsic<[llvm_i32_ty], [llvm_i16_ty, llvm_i16_ty],
                         [IntrNoMem]>;
Index: lib/Target/X86/X86InstrAVX512.td
===================================================================
--- lib/Target/X86/X86InstrAVX512.td
+++ lib/Target/X86/X86InstrAVX512.td
@@ -2295,7 +2295,8 @@
 
 // Mask unpacking
 multiclass avx512_mask_unpck<string Suffix,RegisterClass KRC, ValueType VT,
-                             RegisterClass KRCSrc, Predicate prd> {
+                             RegisterClass KRCSrc, RegisterClass GRC,
+                             Predicate prd> {
   let Predicates = [prd] in {
     def rr : I<0x4b, MRMSrcReg, (outs KRC:$dst),
                (ins KRC:$src1, KRC:$src2),
@@ -2306,22 +2307,20 @@
               (!cast<Instruction>(NAME##rr)
                         (COPY_TO_REGCLASS KRCSrc:$src2, KRC),
                         (COPY_TO_REGCLASS KRCSrc:$src1, KRC))>;
+
+    // Intrinsic call
+    def : Pat<(!cast<Intrinsic>("int_x86_avx512_kunpck_"#Suffix)
+                  GRC:$src1,  GRC:$src2),
+              (COPY_TO_REGCLASS (!cast<Instruction>(NAME##rr)
+                      (VT (COPY_TO_REGCLASS GRC:$src1, KRC)),
+                      (VT (COPY_TO_REGCLASS GRC:$src2, KRC))),
+                  GRC)>;
   }
 }
 
-defm KUNPCKBW : avx512_mask_unpck<"bw", VK16, v16i1, VK8, HasAVX512>, PD;
-defm KUNPCKWD : avx512_mask_unpck<"wd", VK32, v32i1, VK16, HasBWI>, PS;
-defm KUNPCKDQ : avx512_mask_unpck<"dq", VK64, v64i1, VK32, HasBWI>, PS, VEX_W;
-
-multiclass avx512_mask_unpck_int<string IntName, string InstName> {
-  let Predicates = [HasAVX512] in
-    def : Pat<(!cast<Intrinsic>("int_x86_avx512_"##IntName##"_bw")
-                (i16 GR16:$src1), (i16 GR16:$src2)),
-              (COPY_TO_REGCLASS (!cast<Instruction>(InstName##"BWrr")
-              (v16i1 (COPY_TO_REGCLASS GR16:$src1, VK16)),
-              (v16i1 (COPY_TO_REGCLASS GR16:$src2, VK16))), GR16)>;
-}
-defm : avx512_mask_unpck_int<"kunpck",  "KUNPCK">;
+defm KUNPCKBW : avx512_mask_unpck<"bw", VK16, v16i1, VK8, GR16, HasAVX512>, PD;
+defm KUNPCKWD : avx512_mask_unpck<"wd", VK32, v32i1, VK16, GR32, HasBWI>, PS;
+defm KUNPCKDQ : avx512_mask_unpck<"dq", VK64, v64i1, VK32, GR64, HasBWI>, PS, VEX_W;
 
 // Mask bit testing
 multiclass avx512_mask_testop<bits<8> opc, string OpcodeStr, RegisterClass KRC,
Index: test/CodeGen/X86/avx512bw-intrinsics.ll
===================================================================
--- test/CodeGen/X86/avx512bw-intrinsics.ll
+++ test/CodeGen/X86/avx512bw-intrinsics.ll
@@ -1267,3 +1267,32 @@
   %res2 = add  <64 x i8> %res, %res1
   ret  <64 x i8> %res2
 }
+
+declare i32 @llvm.x86.avx512.kunpck.wd(i32, i32)
+
+define i32@test_int_x86_avx512_kunpck_wd(i32 %x0, i32 %x1) {
+; CHECK-LABEL: test_int_x86_avx512_kunpck_wd:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovd %esi, %k0 
+; CHECK-NEXT:    kmovd %edi, %k1 
+; CHECK-NEXT:    kunpckwd %k0, %k1, %k0 
+; CHECK-NEXT:    kmovd %k0, %eax 
+; CHECK-NEXT:    retq 
+  %res = call i32 @llvm.x86.avx512.kunpck.wd(i32 %x0, i32 %x1)
+  ret i32 %res
+}
+
+declare i64 @llvm.x86.avx512.kunpck.dq(i64, i64)
+
+define i64@test_int_x86_avx512_kunpck_qd(i64 %x0, i64 %x1) {
+; CHECK-LABEL: test_int_x86_avx512_kunpck_qd:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovq %rsi, %k0 
+; CHECK-NEXT:    kmovq %rdi, %k1 
+; CHECK-NEXT:    kunpckdq %k0, %k1, %k0 
+; CHECK-NEXT:    kmovq %k0, %rax 
+; CHECK-NEXT:    retq 
+  %res = call i64 @llvm.x86.avx512.kunpck.dq(i64 %x0, i64 %x1)
+  ret i64 %res
+}
+