[AVX512] add intrinsics for integers min/max

[AsafBadouh]

Summary

add intrinsics for min/max with integers 8/16/32/64 bits

[delena]

LGTM

[igorb]

LGTM

include/llvm/IR/IntrinsicsX86.td

def int_x86_avx2_pmins_b : GCCBuiltin<"__builtin_ia32_pminsb256">,
               Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty,
                          llvm_v32i8_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx2_pmins_w : GCCBuiltin<"__builtin_ia32_pminsw256">,
               Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty,
                          llvm_v16i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx2_pmins_d : GCCBuiltin<"__builtin_ia32_pminsd256">,
               Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty,
                          llvm_v8i32_ty], [IntrNoMem, Commutative]>;
+  def int_x86_avx512_mask_pmaxs_b_128 : GCCBuiltin<"__builtin_ia32_pmaxsb128_mask">,
+              Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty, 
+                         llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_b_256 : GCCBuiltin<"__builtin_ia32_pmaxsb256_mask">,
+              Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, 
+                         llvm_v32i8_ty,  llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_b_512 : GCCBuiltin<"__builtin_ia32_pmaxsb512_mask">,
+              Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty, 
+                         llvm_v64i8_ty,  llvm_i64_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_b_128 : GCCBuiltin<"__builtin_ia32_pmaxub128_mask">,
+              Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty, 
+                         llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_b_256 : GCCBuiltin<"__builtin_ia32_pmaxub256_mask">,
+              Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, 
+                         llvm_v32i8_ty,  llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_b_512 : GCCBuiltin<"__builtin_ia32_pmaxub512_mask">,
+              Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty, 
+                         llvm_v64i8_ty,  llvm_i64_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_w_128 : GCCBuiltin<"__builtin_ia32_pmaxsw128_mask">,
+              Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, 
+                         llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_w_256 : GCCBuiltin<"__builtin_ia32_pmaxsw256_mask">,
+              Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, 
+                         llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_w_512 : GCCBuiltin<"__builtin_ia32_pmaxsw512_mask">,
+              Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty, 
+                   llvm_v32i16_ty,  llvm_i32_ty],[IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_w_128 : GCCBuiltin<"__builtin_ia32_pmaxuw128_mask">,
+              Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, 
+                         llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_w_256 : GCCBuiltin<"__builtin_ia32_pmaxuw256_mask">,
+              Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, 
+                         llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_w_512 : GCCBuiltin<"__builtin_ia32_pmaxuw512_mask">,
+              Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty,
+                         llvm_v32i16_ty,  llvm_i32_ty],[IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_b_128 : GCCBuiltin<"__builtin_ia32_pminsb128_mask">,
+              Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty, 
+                         llvm_v16i8_ty,llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_b_256 : GCCBuiltin<"__builtin_ia32_pminsb256_mask">,
+              Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, 
+                         llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_b_512 : GCCBuiltin<"__builtin_ia32_pminsb512_mask">,
+              Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty,
+                         llvm_v64i8_ty, llvm_i64_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_b_128 : GCCBuiltin<"__builtin_ia32_pminub128_mask">,
+              Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty,
+                         llvm_v16i8_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_b_256 : GCCBuiltin<"__builtin_ia32_pminub256_mask">,
+              Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty, llvm_v32i8_ty, 
+                         llvm_v32i8_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_b_512 : GCCBuiltin<"__builtin_ia32_pminub512_mask">,
+              Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty, 
+                         llvm_v64i8_ty,  llvm_i64_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_w_128 : GCCBuiltin<"__builtin_ia32_pminsw128_mask">,
+              Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, 
+                         llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_w_256 : GCCBuiltin<"__builtin_ia32_pminsw256_mask">,
+              Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, 
+                         llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_w_512 : GCCBuiltin<"__builtin_ia32_pminsw512_mask">,
+              Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty,
+                         llvm_v32i16_ty,  llvm_i32_ty],[IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_w_128 : GCCBuiltin<"__builtin_ia32_pminuw128_mask">,
+              Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty, llvm_v8i16_ty, 
+                         llvm_v8i16_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_w_256 : GCCBuiltin<"__builtin_ia32_pminuw256_mask">,
+              Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty, llvm_v16i16_ty, 
+                         llvm_v16i16_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_w_512 : GCCBuiltin<"__builtin_ia32_pminuw512_mask">,
+              Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty, 
+                         llvm_v32i16_ty,  llvm_i32_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pmaxu_d_512 : GCCBuiltin<"__builtin_ia32_pmaxud512_mask">,
               Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
                          llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_d_256 : GCCBuiltin<"__builtin_ia32_pmaxud256_mask">,
+              Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty,
+                         llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_d_128 : GCCBuiltin<"__builtin_ia32_pmaxud128_mask">,
+              Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty,
+                         llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pmaxs_d_512 : GCCBuiltin<"__builtin_ia32_pmaxsd512_mask">,
               Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
                          llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_d_256 : GCCBuiltin<"__builtin_ia32_pmaxsd256_mask">,
+              Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty,
+                         llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_d_128 : GCCBuiltin<"__builtin_ia32_pmaxsd128_mask">,
+              Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty,
+                         llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pmaxu_q_512 : GCCBuiltin<"__builtin_ia32_pmaxuq512_mask">,
               Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
                          llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_q_256 : GCCBuiltin<"__builtin_ia32_pmaxuq256_mask">,
+              Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty,
+                         llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxu_q_128 : GCCBuiltin<"__builtin_ia32_pmaxuq128_mask">,
+              Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty,
+                         llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pmaxs_q_512 : GCCBuiltin<"__builtin_ia32_pmaxsq512_mask">,
               Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
                          llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_q_256 : GCCBuiltin<"__builtin_ia32_pmaxsq256_mask">,
+              Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty,
+                         llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmaxs_q_128 : GCCBuiltin<"__builtin_ia32_pmaxsq128_mask">,
+              Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty,
+                         llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pminu_d_512 : GCCBuiltin<"__builtin_ia32_pminud512_mask">,
               Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
                          llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_d_256 : GCCBuiltin<"__builtin_ia32_pminud256_mask">,
+              Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty,
+                         llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_d_128 : GCCBuiltin<"__builtin_ia32_pminud128_mask">,
+              Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty,
+                         llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pmins_d_512 : GCCBuiltin<"__builtin_ia32_pminsd512_mask">,
               Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
                          llvm_v16i32_ty, llvm_i16_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_d_256 : GCCBuiltin<"__builtin_ia32_pminsd256_mask">,
+              Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty,
+                         llvm_v8i32_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_d_128 : GCCBuiltin<"__builtin_ia32_pminsd128_mask">,
+              Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty,
+                         llvm_v4i32_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pminu_q_512 : GCCBuiltin<"__builtin_ia32_pminuq512_mask">,
               Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
                          llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_q_256 : GCCBuiltin<"__builtin_ia32_pminuq256_mask">,
+              Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty,
+                         llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pminu_q_128 : GCCBuiltin<"__builtin_ia32_pminuq128_mask">,
+              Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty,
+                         llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>;
   def int_x86_avx512_mask_pmins_q_512 : GCCBuiltin<"__builtin_ia32_pminsq512_mask">,
               Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
                          llvm_v8i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_q_256 : GCCBuiltin<"__builtin_ia32_pminsq256_mask">,
+              Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty,
+                         llvm_v4i64_ty, llvm_i8_ty], [IntrNoMem]>;
+  def int_x86_avx512_mask_pmins_q_128 : GCCBuiltin<"__builtin_ia32_pminsq128_mask">,
+              Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty,
+                         llvm_v2i64_ty, llvm_i8_ty], [IntrNoMem]>;
 }
 
 // Integer shift ops.
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
   def int_x86_avx2_psll_w : GCCBuiltin<"__builtin_ia32_psllw256">,
               Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty,
                          llvm_v8i16_ty], [IntrNoMem]>;
   def int_x86_avx2_psll_d : GCCBuiltin<"__builtin_ia32_pslld256">,

lib/Target/X86/X86InstrAVX512.td

 
 defm VPMINUB : avx512_binop_rm_vl_b<0xDA, "vpminu", X86umin,
                                      SSE_INTALU_ITINS_P, HasBWI, 1>;
 defm VPMINUW : avx512_binop_rm_vl_w<0x3A, "vpminu", X86umin,
                                      SSE_INTALU_ITINS_P, HasBWI, 1>, T8PD;
 defm VPMINU : avx512_binop_rm_vl_dq<0x3B, 0x3B, "vpminu", X86umin,
                                      SSE_INTALU_ITINS_P, HasAVX512, 1>, T8PD;
 
-def : Pat <(v16i32 (int_x86_avx512_mask_pmaxs_d_512 (v16i32 VR512:$src1),
-                    (v16i32 VR512:$src2), (v16i32 immAllZerosV), (i16 -1))),
-           (VPMAXSDZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v16i32 (int_x86_avx512_mask_pmaxu_d_512 (v16i32 VR512:$src1),
-                    (v16i32 VR512:$src2), (v16i32 immAllZerosV), (i16 -1))),
-           (VPMAXUDZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v8i64 (int_x86_avx512_mask_pmaxs_q_512 (v8i64 VR512:$src1),
-                (v8i64 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
-           (VPMAXSQZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v8i64 (int_x86_avx512_mask_pmaxu_q_512 (v8i64 VR512:$src1),
-                (v8i64 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
-           (VPMAXUQZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v16i32 (int_x86_avx512_mask_pmins_d_512 (v16i32 VR512:$src1),
-                    (v16i32 VR512:$src2), (v16i32 immAllZerosV), (i16 -1))),
-           (VPMINSDZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v16i32 (int_x86_avx512_mask_pminu_d_512 (v16i32 VR512:$src1),
-                    (v16i32 VR512:$src2), (v16i32 immAllZerosV), (i16 -1))),
-           (VPMINUDZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v8i64 (int_x86_avx512_mask_pmins_q_512 (v8i64 VR512:$src1),
-                (v8i64 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
-           (VPMINSQZrr VR512:$src1, VR512:$src2)>;
-def : Pat <(v8i64 (int_x86_avx512_mask_pminu_q_512 (v8i64 VR512:$src1),
-                (v8i64 VR512:$src2), (bc_v8i64 (v16i32 immAllZerosV)), (i8 -1))),
-           (VPMINUQZrr VR512:$src1, VR512:$src2)>;
 //===----------------------------------------------------------------------===//
 // AVX-512 - Unpack Instructions
 //===----------------------------------------------------------------------===//
 
 multiclass avx512_unpack_fp<bits<8> opc, SDNode OpNode, ValueType vt,
                                    PatFrag mem_frag, RegisterClass RC,
                                    X86MemOperand x86memop, string asm,
                                    Domain d> {

lib/Target/X86/X86IntrinsicsInfo.h

static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_d_256,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_d_512,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_q_128,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_q_256,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   X86_INTRINSIC_DATA(avx512_mask_pcmpgt_q_512,  CMP_MASK,  X86ISD::PCMPGTM, 0),
   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_pmaxs_b_128, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_b_256, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_b_512, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_d_128, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_d_256, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_d_512, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_q_128, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_q_256, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_q_512, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_w_128, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_w_256, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxs_w_512, INTR_TYPE_2OP_MASK, X86ISD::SMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_b_128, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_b_256, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_b_512, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_d_128, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_d_256, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_d_512, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_q_128, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_q_256, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_q_512, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_w_128, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_w_256, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmaxu_w_512, INTR_TYPE_2OP_MASK, X86ISD::UMAX, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_b_128, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_b_256, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_b_512, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_d_128, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_d_256, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_d_512, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_q_128, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_q_256, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_q_512, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_w_128, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_w_256, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pmins_w_512, INTR_TYPE_2OP_MASK, X86ISD::SMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_b_128, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_b_256, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_b_512, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_d_128, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_d_256, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_d_512, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_q_128, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_q_256, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_q_512, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_w_128, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_w_256, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
+  X86_INTRINSIC_DATA(avx512_mask_pminu_w_512, INTR_TYPE_2OP_MASK, X86ISD::UMIN, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmul_dq_128, INTR_TYPE_2OP_MASK,
                      X86ISD::PMULDQ, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmul_dq_256, INTR_TYPE_2OP_MASK,
                      X86ISD::PMULDQ, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmul_dq_512, INTR_TYPE_2OP_MASK,
                      X86ISD::PMULDQ, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmull_d_128, INTR_TYPE_2OP_MASK, ISD::MUL, 0),
   X86_INTRINSIC_DATA(avx512_mask_pmull_d_256, INTR_TYPE_2OP_MASK, ISD::MUL, 0),

test/CodeGen/X86/avx512-intrinsics.ll

 
  define <8 x i64> @test_pabsq(<8 x i64> %a) {
  ;CHECK: vpabsq {{.*}}encoding: [0x62,0xf2,0xfd,0x48,0x1f,0xc0]
  %res = call <8 x i64> @llvm.x86.avx512.mask.pabs.q.512(<8 x i64> %a, <8 x i64>zeroinitializer, i8 -1)
  ret <8 x i64> %res
  }
  declare <8 x i64> @llvm.x86.avx512.mask.pabs.q.512(<8 x i64>, <8 x i64>, i8)
 
-define <8 x i64> @test_vpmaxq(<8 x i64> %a0, <8 x i64> %a1) {
-  ; CHECK: vpmaxsq {{.*}}encoding: [0x62,0xf2,0xfd,0x48,0x3d,0xc1]
-  %res = call <8 x i64> @llvm.x86.avx512.mask.pmaxs.q.512(<8 x i64> %a0, <8 x i64> %a1,
-                    <8 x i64>zeroinitializer, i8 -1)
-  ret <8 x i64> %res
-}
-declare <8 x i64> @llvm.x86.avx512.mask.pmaxs.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
-
-define <16 x i32> @test_vpminud(<16 x i32> %a0, <16 x i32> %a1) {
-  ; CHECK: vpminud {{.*}}encoding: [0x62,0xf2,0x7d,0x48,0x3b,0xc1]
-  %res = call <16 x i32> @llvm.x86.avx512.mask.pminu.d.512(<16 x i32> %a0, <16 x i32> %a1,
-                    <16 x i32>zeroinitializer, i16 -1)
-  ret <16 x i32> %res
-}
-declare <16 x i32> @llvm.x86.avx512.mask.pminu.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
-
-define <16 x i32> @test_vpmaxsd(<16 x i32> %a0, <16 x i32> %a1) {
-  ; CHECK: vpmaxsd {{.*}}encoding: [0x62,0xf2,0x7d,0x48,0x3d,0xc1]
-  %res = call <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32> %a0, <16 x i32> %a1,
-                    <16 x i32>zeroinitializer, i16 -1)
-  ret <16 x i32> %res
-}
-declare <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
-
 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)
   ret i8 %res
 }
 declare i8 @llvm.x86.avx512.mask.ptestm.q.512(<8 x i64>, <8 x i64>, i8)
 
 define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1) {

 }
 
 define <2 x double> @test_max_sd(<2 x double> %a0, <2 x double> %a1) {
 ; CHECK-LABEL: test_max_sd
 ; CHECK: vmaxsd  %xmm1, %xmm0, %xmm0
   %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 4)
   ret <2 x double> %res
 }
+
+define <8 x i64> @test_vpmaxq(<8 x i64> %a0, <8 x i64> %a1) {
+  ; CHECK: vpmaxsq {{.*}}encoding: [0x62,0xf2,0xfd,0x48,0x3d,0xc1]
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmaxs.q.512(<8 x i64> %a0, <8 x i64> %a1,
+                    <8 x i64>zeroinitializer, i8 -1)
+  ret <8 x i64> %res
+}
+declare <8 x i64> @llvm.x86.avx512.mask.pmaxs.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
+
+define <16 x i32> @test_vpminud(<16 x i32> %a0, <16 x i32> %a1) {
+  ; CHECK: vpminud {{.*}}encoding: [0x62,0xf2,0x7d,0x48,0x3b,0xc1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pminu.d.512(<16 x i32> %a0, <16 x i32> %a1,
+                    <16 x i32>zeroinitializer, i16 -1)
+  ret <16 x i32> %res
+}
+declare <16 x i32> @llvm.x86.avx512.mask.pminu.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+define <16 x i32> @test_vpmaxsd(<16 x i32> %a0, <16 x i32> %a1) {
+  ; CHECK: vpmaxsd {{.*}}encoding: [0x62,0xf2,0x7d,0x48,0x3d,0xc1]
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32> %a0, <16 x i32> %a1,
+                    <16 x i32>zeroinitializer, i16 -1)
+  ret <16 x i32> %res
+}
+declare <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_d_512
+; CHECK-NOT: call 
+; CHECK: vpmaxsd %zmm
+; CHECK: {%k1} 
+define <16 x i32>@test_int_x86_avx512_mask_pmaxs_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
+  %res1 = call <16 x i32> @llvm.x86.avx512.mask.pmaxs.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
+  %res2 = add <16 x i32> %res, %res1
+  ret <16 x i32> %res2
+}
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_q_512
+; CHECK-NOT: call 
+; CHECK: vpmaxsq %zmm
+; CHECK: {%k1} 
+define <8 x i64>@test_int_x86_avx512_mask_pmaxs_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmaxs.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  %res1 = call <8 x i64> @llvm.x86.avx512.mask.pmaxs.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
+  %res2 = add <8 x i64> %res, %res1
+  ret <8 x i64> %res2
+}
+
+declare <16 x i32> @llvm.x86.avx512.mask.pmaxu.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_d_512
+; CHECK-NOT: call 
+; CHECK: vpmaxud %zmm
+; CHECK: {%k1} 
+define <16 x i32>@test_int_x86_avx512_mask_pmaxu_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pmaxu.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
+  %res1 = call <16 x i32> @llvm.x86.avx512.mask.pmaxu.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
+  %res2 = add <16 x i32> %res, %res1
+  ret <16 x i32> %res2
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.pmaxu.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_q_512
+; CHECK-NOT: call 
+; CHECK: vpmaxuq %zmm
+; CHECK: {%k1} 
+define <8 x i64>@test_int_x86_avx512_mask_pmaxu_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmaxu.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  %res1 = call <8 x i64> @llvm.x86.avx512.mask.pmaxu.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
+  %res2 = add <8 x i64> %res, %res1
+  ret <8 x i64> %res2
+}
+
+declare <16 x i32> @llvm.x86.avx512.mask.pmins.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_d_512
+; CHECK-NOT: call 
+; CHECK: vpminsd %zmm
+; CHECK: {%k1} 
+define <16 x i32>@test_int_x86_avx512_mask_pmins_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pmins.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
+  %res1 = call <16 x i32> @llvm.x86.avx512.mask.pmins.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
+  %res2 = add <16 x i32> %res, %res1
+  ret <16 x i32> %res2
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.pmins.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_q_512
+; CHECK-NOT: call 
+; CHECK: vpminsq %zmm
+; CHECK: {%k1} 
+define <8 x i64>@test_int_x86_avx512_mask_pmins_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pmins.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  %res1 = call <8 x i64> @llvm.x86.avx512.mask.pmins.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
+  %res2 = add <8 x i64> %res, %res1
+  ret <8 x i64> %res2
+}
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_d_512
+; CHECK-NOT: call 
+; CHECK: vpminud %zmm
+; CHECK: {%k1} 
+define <16 x i32>@test_int_x86_avx512_mask_pminu_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
+  %res = call <16 x i32> @llvm.x86.avx512.mask.pminu.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
+  %res1 = call <16 x i32> @llvm.x86.avx512.mask.pminu.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
+  %res2 = add <16 x i32> %res, %res1
+  ret <16 x i32> %res2
+}
+
+declare <8 x i64> @llvm.x86.avx512.mask.pminu.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_q_512
+; CHECK-NOT: call 
+; CHECK: vpminuq %zmm
+; CHECK: {%k1} 
+define <8 x i64>@test_int_x86_avx512_mask_pminu_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
+  %res = call <8 x i64> @llvm.x86.avx512.mask.pminu.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
+  %res1 = call <8 x i64> @llvm.x86.avx512.mask.pminu.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
+  %res2 = add <8 x i64> %res, %res1
+  ret <8 x i64> %res2
+}

test/CodeGen/X86/avx512bw-intrinsics.ll

define <32 x i16> @test_mask_subs_epu16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
   ;CHECK-LABEL: test_mask_subs_epu16_rmkz_512
   ;CHECK: vpsubusw (%rdi), %zmm0, %zmm0 {%k1} {z} 
   %b = load <32 x i16>, <32 x i16>* %ptr_b
   %res = call <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 %mask)
   ret <32 x i16> %res
 }
 
 declare <32 x i16> @llvm.x86.avx512.mask.psubus.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+declare <64 x i8> @llvm.x86.avx512.mask.pmaxs.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_b_512
+; CHECK-NOT: call 
+; CHECK: vpmaxsb %zmm
+; CHECK: {%k1} 
+define <64 x i8>@test_int_x86_avx512_mask_pmaxs_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmaxs.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  %res1 = call <64 x i8> @llvm.x86.avx512.mask.pmaxs.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  %res2 = add <64 x i8> %res, %res1
+  ret <64 x i8> %res2
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmaxs.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_w_512
+; CHECK-NOT: call 
+; CHECK: vpmaxsw %zmm
+; CHECK: {%k1} 
+define <32 x i16>@test_int_x86_avx512_mask_pmaxs_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaxs.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.pmaxs.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  %res2 = add <32 x i16> %res, %res1
+  ret <32 x i16> %res2
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pmaxu.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_b_512
+; CHECK-NOT: call 
+; CHECK: vpmaxub %zmm
+; CHECK: {%k1} 
+define <64 x i8>@test_int_x86_avx512_mask_pmaxu_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmaxu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  %res1 = call <64 x i8> @llvm.x86.avx512.mask.pmaxu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  %res2 = add <64 x i8> %res, %res1
+  ret <64 x i8> %res2
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmaxu.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_w_512
+; CHECK-NOT: call 
+; CHECK: vpmaxuw %zmm
+; CHECK: {%k1} 
+define <32 x i16>@test_int_x86_avx512_mask_pmaxu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmaxu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.pmaxu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  %res2 = add <32 x i16> %res, %res1
+  ret <32 x i16> %res2
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pmins.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_b_512
+; CHECK-NOT: call 
+; CHECK: vpminsb %zmm
+; CHECK: {%k1} 
+define <64 x i8>@test_int_x86_avx512_mask_pmins_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pmins.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  %res1 = call <64 x i8> @llvm.x86.avx512.mask.pmins.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  %res2 = add <64 x i8> %res, %res1
+  ret <64 x i8> %res2
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pmins.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_w_512
+; CHECK-NOT: call 
+; CHECK: vpminsw %zmm
+; CHECK: {%k1} 
+define <32 x i16>@test_int_x86_avx512_mask_pmins_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pmins.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.pmins.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  %res2 = add <32 x i16> %res, %res1
+  ret <32 x i16> %res2
+}
+
+declare <64 x i8> @llvm.x86.avx512.mask.pminu.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_b_512
+; CHECK-NOT: call 
+; CHECK: vpminub %zmm
+; CHECK: {%k1} 
+define <64 x i8>@test_int_x86_avx512_mask_pminu_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
+  %res = call <64 x i8> @llvm.x86.avx512.mask.pminu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3)
+  %res1 = call <64 x i8> @llvm.x86.avx512.mask.pminu.b.512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 -1)
+  %res2 = add <64 x i8> %res, %res1
+  ret <64 x i8> %res2
+}
+
+declare <32 x i16> @llvm.x86.avx512.mask.pminu.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_w_512
+; CHECK-NOT: call 
+; CHECK: vpminuw %zmm
+; CHECK: {%k1} 
+define <32 x i16>@test_int_x86_avx512_mask_pminu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
+  %res = call <32 x i16> @llvm.x86.avx512.mask.pminu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3)
+  %res1 = call <32 x i16> @llvm.x86.avx512.mask.pminu.w.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 -1)
+  %res2 = add <32 x i16> %res, %res1
+  ret <32 x i16> %res2
+}

test/CodeGen/X86/avx512bwvl-intrinsics.ll

define <4 x float> @test_mask_vfmadd128_ps_rz(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rz
   ; CHECK: vfmadd213ps %xmm2, %xmm1, %xmm0 ## encoding: [0x62,0xf2,0x75,0x08,0xa8,0xc2]
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmk(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmk
-  ; CHECK: vfmadd213ps  (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmka(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmka
   ; CHECK: vfmadd213ps     (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2, align 8
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmkz(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmkz
-  ; CHECK: vfmadd213ps  (%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0x07]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmkza(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmkza
-  ; CHECK: vfmadd213ps  (%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0x07]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2, align 4
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmb(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmb
-  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
   %q = load float, float* %ptr_a2
   %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 %mask) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmba(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmba
-  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
   %q = load float, float* %ptr_a2, align 4
   %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 %mask) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmbz(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmbz
-  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0  ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0  ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
   %q = load float, float* %ptr_a2
   %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 -1) nounwind
   ret <4 x float> %res
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmbza(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_ps_rmbza
-  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0  ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
+  ; CHECK: vfmadd213ps  (%rdi){1to4}, %xmm1, %xmm0  ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
   %q = load float, float* %ptr_a2, align 4
   %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
   %res = call <4 x float> @llvm.x86.fma.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 -1) nounwind
   ret <4 x float> %res
 }
define <2 x double> @test_mask_vfmadd128_pd_rz(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_pd_rz
   ; CHECK: vfmadd213pd %xmm2, %xmm1, %xmm0 ## encoding: [0x62,0xf2,0xf5,0x08,0xa8,0xc2]
   %res = call <2 x double> @llvm.x86.fma.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
   ret <2 x double> %res
 }
 
 define <2 x double> @test_mask_vfmadd128_pd_rmk(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd128_pd_rmk
-  ; CHECK: vfmadd213pd  (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa8,0x07]
+  ; CHECK: vfmadd213pd  (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa8,0x07]
   %a2 = load <2 x double>, <2 x double>* %ptr_a2
   %res = call <2 x double> @llvm.x86.fma.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
   ret <2 x double> %res
 }
 
 define <2 x double> @test_mask_vfmadd128_pd_rmkz(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2) {
   ; CHECK-LABEL: test_mask_vfmadd128_pd_rmkz
-  ; CHECK: vfmadd213pd  (%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0x07]
+  ; CHECK: vfmadd213pd  (%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0x07]
   %a2 = load <2 x double>, <2 x double>* %ptr_a2
   %res = call <2 x double> @llvm.x86.fma.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
   ret <2 x double> %res
 }
 
 define <4 x double> @test_mask_vfmadd256_pd_r(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd256_pd_r
   ; CHECK: vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa8,0xc2]
   %res = call <4 x double> @llvm.x86.fma.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
   ret <4 x double> %res
 }
 
 define <4 x double> @test_mask_vfmadd256_pd_rz(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
   ; CHECK-LABEL: test_mask_vfmadd256_pd_rz
   ; CHECK: vfmadd213pd %ymm2, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0xf5,0x28,0xa8,0xc2]
   %res = call <4 x double> @llvm.x86.fma.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
   ret <4 x double> %res
 }
 
 define <4 x double> @test_mask_vfmadd256_pd_rmk(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2, i8 %mask) {
   ; CHECK-LABEL: test_mask_vfmadd256_pd_rmk
-  ; CHECK: vfmadd213pd  (%rdi), %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa8,0x07]
+  ; CHECK: vfmadd213pd  (%rdi), %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa8,0x07]
   %a2 = load <4 x double>, <4 x double>* %ptr_a2
   %res = call <4 x double> @llvm.x86.fma.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
   ret <4 x double> %res
 }
 
 define <4 x double> @test_mask_vfmadd256_pd_rmkz(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2) {
   ; CHECK-LABEL: test_mask_vfmadd256_pd_rmkz
-  ; CHECK: vfmadd213pd  (%rdi), %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0x07]
+  ; CHECK: vfmadd213pd  (%rdi), %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0x07]
   %a2 = load <4 x double>, <4 x double>* %ptr_a2
   %res = call <4 x double> @llvm.x86.fma.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
   ret <4 x double> %res
 }
 define <8 x i16> @test_mask_add_epi16_rr_128(<8 x i16> %a, <8 x i16> %b) {
   ;CHECK-LABEL: test_mask_add_epi16_rr_128
   ;CHECK: vpaddw %xmm1, %xmm0, %xmm0     ## encoding: [0x62,0xf1,0x7d,0x08,0xfd,0xc1]
   %res = call <8 x i16> @llvm.x86.avx512.mask.padd.w.128(<8 x i16> %a, <8 x i16> %b, <8 x i16> zeroinitializer, i8 -1)

 define <32 x i8> @test_mask_subs_epu8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
   ;CHECK-LABEL: test_mask_subs_epu8_rmkz_256
   ;CHECK: vpsubusb (%rdi), %ymm0, %ymm0 {%k1} {z} 
   %b = load <32 x i8>, <32 x i8>* %ptr_b
   %res = call <32 x i8> @llvm.x86.avx512.mask.psubus.b.256(<32 x i8> %a, <32 x i8> %b, <32 x i8> zeroinitializer, i32 %mask)
   ret <32 x i8> %res
 }
 
-declare <32 x i8> @llvm.x86.avx512.mask.psubus.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
+declare <32 x i8> @llvm.x86.avx512.mask.psubus.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
+
+declare <16 x i8> @llvm.x86.avx512.mask.pmaxs.b.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_b_128
+; CHECK-NOT: call 
+; CHECK: vpmaxsb %xmm
+; CHECK: {%k1} 
+define <16 x i8>@test_int_x86_avx512_mask_pmaxs_b_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %mask) {
+  %res = call <16 x i8> @llvm.x86.avx512.mask.pmaxs.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2 ,i16 %mask)
+  %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmaxs.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i8> %res, %res1
+  ret <16 x i8> %res2
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pmaxs.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_b_256
+; CHECK-NOT: call 
+; CHECK: vpmaxsb %ymm
+; CHECK: {%k1} 
+define <32 x i8>@test_int_x86_avx512_mask_pmaxs_b_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmaxs.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
+  %res1 = call <32 x i8> @llvm.x86.avx512.mask.pmaxs.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
+  %res2 = add <32 x i8> %res, %res1
+  ret <32 x i8> %res2
+}
+
+declare <8 x i16> @llvm.x86.avx512.mask.pmaxs.w.128(<8 x i16>, <8 x i16>, <8 x i16>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_w_128
+; CHECK-NOT: call 
+; CHECK: vpmaxsw %xmm
+; CHECK: {%k1} 
+define <8 x i16>@test_int_x86_avx512_mask_pmaxs_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
+  %res = call <8 x i16> @llvm.x86.avx512.mask.pmaxs.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3)
+  %res1 = call <8 x i16> @llvm.x86.avx512.mask.pmaxs.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 -1)
+  %res2 = add <8 x i16> %res, %res1
+  ret <8 x i16> %res2
+}
+
+declare <16 x i16> @llvm.x86.avx512.mask.pmaxs.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_w_256
+; CHECK-NOT: call 
+; CHECK: vpmaxsw %ymm
+; CHECK: {%k1} 
+define <16 x i16>@test_int_x86_avx512_mask_pmaxs_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask) {
+  %res = call <16 x i16> @llvm.x86.avx512.mask.pmaxs.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask)
+  %res1 = call <16 x i16> @llvm.x86.avx512.mask.pmaxs.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i16> %res, %res1
+  ret <16 x i16> %res2
+}
+
+declare <16 x i8> @llvm.x86.avx512.mask.pmaxu.b.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_b_128
+; CHECK-NOT: call 
+; CHECK: vpmaxub %xmm
+; CHECK: {%k1} 
+define <16 x i8>@test_int_x86_avx512_mask_pmaxu_b_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2,i16 %mask) {
+  %res = call <16 x i8> @llvm.x86.avx512.mask.pmaxu.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %mask)
+  %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmaxu.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i8> %res, %res1
+  ret <16 x i8> %res2
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pmaxu.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_b_256
+; CHECK-NOT: call 
+; CHECK: vpmaxub %ymm
+; CHECK: {%k1} 
+define <32 x i8>@test_int_x86_avx512_mask_pmaxu_b_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmaxu.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
+  %res1 = call <32 x i8> @llvm.x86.avx512.mask.pmaxu.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
+  %res2 = add <32 x i8> %res, %res1
+  ret <32 x i8> %res2
+}
+
+declare <8 x i16> @llvm.x86.avx512.mask.pmaxu.w.128(<8 x i16>, <8 x i16>, <8 x i16>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_w_128
+; CHECK-NOT: call 
+; CHECK: vpmaxuw %xmm
+; CHECK: {%k1} 
+define <8 x i16>@test_int_x86_avx512_mask_pmaxu_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
+  %res = call <8 x i16> @llvm.x86.avx512.mask.pmaxu.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3)
+  %res1 = call <8 x i16> @llvm.x86.avx512.mask.pmaxu.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 -1)
+  %res2 = add <8 x i16> %res, %res1
+  ret <8 x i16> %res2
+}
+
+declare <16 x i16> @llvm.x86.avx512.mask.pmaxu.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_w_256
+; CHECK-NOT: call 
+; CHECK: vpmaxuw %ymm
+; CHECK: {%k1} 
+define <16 x i16>@test_int_x86_avx512_mask_pmaxu_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask) {
+  %res = call <16 x i16> @llvm.x86.avx512.mask.pmaxu.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask)
+  %res1 = call <16 x i16> @llvm.x86.avx512.mask.pmaxu.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i16> %res, %res1
+  ret <16 x i16> %res2
+}
+
+declare <16 x i8> @llvm.x86.avx512.mask.pmins.b.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_b_128
+; CHECK-NOT: call 
+; CHECK: vpminsb %xmm
+; CHECK: {%k1} 
+define <16 x i8>@test_int_x86_avx512_mask_pmins_b_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %mask) {
+  %res = call <16 x i8> @llvm.x86.avx512.mask.pmins.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %mask)
+  %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmins.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i8> %res, %res1
+  ret <16 x i8> %res2
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pmins.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_b_256
+; CHECK-NOT: call 
+; CHECK: vpminsb %ymm
+; CHECK: {%k1} 
+define <32 x i8>@test_int_x86_avx512_mask_pmins_b_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pmins.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
+  %res1 = call <32 x i8> @llvm.x86.avx512.mask.pmins.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
+  %res2 = add <32 x i8> %res, %res1
+  ret <32 x i8> %res2
+}
+
+declare <8 x i16> @llvm.x86.avx512.mask.pmins.w.128(<8 x i16>, <8 x i16>, <8 x i16>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_w_128
+; CHECK-NOT: call 
+; CHECK: vpminsw %xmm
+; CHECK: {%k1} 
+define <8 x i16>@test_int_x86_avx512_mask_pmins_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
+  %res = call <8 x i16> @llvm.x86.avx512.mask.pmins.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3)
+  %res1 = call <8 x i16> @llvm.x86.avx512.mask.pmins.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 -1)
+  %res2 = add <8 x i16> %res, %res1
+  ret <8 x i16> %res2
+}
+
+declare <16 x i16> @llvm.x86.avx512.mask.pmins.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_w_256
+; CHECK-NOT: call 
+; CHECK: vpminsw %ymm
+; CHECK: {%k1} 
+define <16 x i16>@test_int_x86_avx512_mask_pmins_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask) {
+  %res = call <16 x i16> @llvm.x86.avx512.mask.pmins.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask)
+  %res1 = call <16 x i16> @llvm.x86.avx512.mask.pmins.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i16> %res, %res1
+  ret <16 x i16> %res2
+}
+
+declare <16 x i8> @llvm.x86.avx512.mask.pminu.b.128(<16 x i8>, <16 x i8>, <16 x i8>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_b_128
+; CHECK-NOT: call 
+; CHECK: vpminub %xmm
+; CHECK: {%k1} 
+define <16 x i8>@test_int_x86_avx512_mask_pminu_b_128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %mask) {
+  %res = call <16 x i8> @llvm.x86.avx512.mask.pminu.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %mask)
+  %res1 = call <16 x i8> @llvm.x86.avx512.mask.pminu.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i8> %res, %res1
+  ret <16 x i8> %res2
+}
+
+declare <32 x i8> @llvm.x86.avx512.mask.pminu.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_b_256
+; CHECK-NOT: call 
+; CHECK: vpminub %ymm
+; CHECK: {%k1} 
+define <32 x i8>@test_int_x86_avx512_mask_pminu_b_256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3) {
+  %res = call <32 x i8> @llvm.x86.avx512.mask.pminu.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)
+  %res1 = call <32 x i8> @llvm.x86.avx512.mask.pminu.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 -1)
+  %res2 = add <32 x i8> %res, %res1
+  ret <32 x i8> %res2
+}
+
+declare <8 x i16> @llvm.x86.avx512.mask.pminu.w.128(<8 x i16>, <8 x i16>, <8 x i16>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_w_128
+; CHECK-NOT: call 
+; CHECK: vpminuw %xmm
+; CHECK: {%k1} 
+define <8 x i16>@test_int_x86_avx512_mask_pminu_w_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
+  %res = call <8 x i16> @llvm.x86.avx512.mask.pminu.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3)
+  %res1 = call <8 x i16> @llvm.x86.avx512.mask.pminu.w.128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 -1)
+  %res2 = add <8 x i16> %res, %res1
+  ret <8 x i16> %res2
+}
+
+declare <16 x i16> @llvm.x86.avx512.mask.pminu.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_w_256
+; CHECK-NOT: call 
+; CHECK: vpminuw %ymm
+; CHECK: {%k1} 
+define <16 x i16>@test_int_x86_avx512_mask_pminu_w_256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask) {
+  %res = call <16 x i16> @llvm.x86.avx512.mask.pminu.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> %x2, i16 %mask)
+  %res1 = call <16 x i16> @llvm.x86.avx512.mask.pminu.w.256(<16 x i16> %x0, <16 x i16> %x1, <16 x i16> zeroinitializer, i16 %mask)
+  %res2 = add <16 x i16> %res, %res1
+  ret <16 x i16> %res2
+}
+

test/CodeGen/X86/avx512vl-intrinsics.ll

 declare <4 x double> @llvm.x86.avx512.mask.getexp.pd.256(<4 x double>, <4 x double>, i8) nounwind readnone
 
 define <8 x float> @test_getexp_ps_256(<8 x float> %a0) {
   ; CHECK-LABEL: test_getexp_ps_256
   ; CHECK: vgetexpps
   %res = call <8 x float> @llvm.x86.avx512.mask.getexp.ps.256(<8 x float> %a0, <8 x float> zeroinitializer, i8 -1)
   ret <8 x float> %res
 }
-declare <8 x float> @llvm.x86.avx512.mask.getexp.ps.256(<8 x float>, <8 x float>, i8) nounwind readnone
+declare <8 x float> @llvm.x86.avx512.mask.getexp.ps.256(<8 x float>, <8 x float>, i8) nounwind readnone
+
+declare <4 x i32> @llvm.x86.avx512.mask.pmaxs.d.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_d_128
+; CHECK-NOT: call 
+; CHECK: vpmaxsd %xmm
+; CHECK: {%k1} 
+define <4 x i32>@test_int_x86_avx512_mask_pmaxs_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %mask) {
+  %res = call <4 x i32> @llvm.x86.avx512.mask.pmaxs.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2 ,i8 %mask)
+  %res1 = call <4 x i32> @llvm.x86.avx512.mask.pmaxs.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i32> %res, %res1
+  ret <4 x i32> %res2
+}
+
+declare <8 x i32> @llvm.x86.avx512.mask.pmaxs.d.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_d_256
+; CHECK-NOT: call 
+; CHECK: vpmaxsd %ymm
+; CHECK: {%k1} 
+define <8 x i32>@test_int_x86_avx512_mask_pmaxs_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3) {
+  %res = call <8 x i32> @llvm.x86.avx512.mask.pmaxs.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
+  %res1 = call <8 x i32> @llvm.x86.avx512.mask.pmaxs.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
+  %res2 = add <8 x i32> %res, %res1
+  ret <8 x i32> %res2
+}
+
+declare <2 x i64> @llvm.x86.avx512.mask.pmaxs.q.128(<2 x i64>, <2 x i64>, <2 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_q_128
+; CHECK-NOT: call 
+; CHECK: vpmaxsq %xmm
+; CHECK: {%k1} 
+define <2 x i64>@test_int_x86_avx512_mask_pmaxs_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
+  %res = call <2 x i64> @llvm.x86.avx512.mask.pmaxs.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3)
+  %res1 = call <2 x i64> @llvm.x86.avx512.mask.pmaxs.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 -1)
+  %res2 = add <2 x i64> %res, %res1
+  ret <2 x i64> %res2
+}
+
+declare <4 x i64> @llvm.x86.avx512.mask.pmaxs.q.256(<4 x i64>, <4 x i64>, <4 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxs_q_256
+; CHECK-NOT: call 
+; CHECK: vpmaxsq %ymm
+; CHECK: {%k1} 
+define <4 x i64>@test_int_x86_avx512_mask_pmaxs_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask) {
+  %res = call <4 x i64> @llvm.x86.avx512.mask.pmaxs.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask)
+  %res1 = call <4 x i64> @llvm.x86.avx512.mask.pmaxs.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i64> %res, %res1
+  ret <4 x i64> %res2
+}
+
+declare <4 x i32> @llvm.x86.avx512.mask.pmaxu.d.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_d_128
+; CHECK-NOT: call 
+; CHECK: vpmaxud %xmm
+; CHECK: {%k1} 
+define <4 x i32>@test_int_x86_avx512_mask_pmaxu_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2,i8 %mask) {
+  %res = call <4 x i32> @llvm.x86.avx512.mask.pmaxu.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %mask)
+  %res1 = call <4 x i32> @llvm.x86.avx512.mask.pmaxu.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i32> %res, %res1
+  ret <4 x i32> %res2
+}
+
+declare <8 x i32> @llvm.x86.avx512.mask.pmaxu.d.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_d_256
+; CHECK-NOT: call 
+; CHECK: vpmaxud %ymm
+; CHECK: {%k1} 
+define <8 x i32>@test_int_x86_avx512_mask_pmaxu_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3) {
+  %res = call <8 x i32> @llvm.x86.avx512.mask.pmaxu.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
+  %res1 = call <8 x i32> @llvm.x86.avx512.mask.pmaxu.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
+  %res2 = add <8 x i32> %res, %res1
+  ret <8 x i32> %res2
+}
+
+declare <2 x i64> @llvm.x86.avx512.mask.pmaxu.q.128(<2 x i64>, <2 x i64>, <2 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_q_128
+; CHECK-NOT: call 
+; CHECK: vpmaxuq %xmm
+; CHECK: {%k1} 
+define <2 x i64>@test_int_x86_avx512_mask_pmaxu_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
+  %res = call <2 x i64> @llvm.x86.avx512.mask.pmaxu.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3)
+  %res1 = call <2 x i64> @llvm.x86.avx512.mask.pmaxu.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 -1)
+  %res2 = add <2 x i64> %res, %res1
+  ret <2 x i64> %res2
+}
+
+declare <4 x i64> @llvm.x86.avx512.mask.pmaxu.q.256(<4 x i64>, <4 x i64>, <4 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmaxu_q_256
+; CHECK-NOT: call 
+; CHECK: vpmaxuq %ymm
+; CHECK: {%k1} 
+define <4 x i64>@test_int_x86_avx512_mask_pmaxu_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask) {
+  %res = call <4 x i64> @llvm.x86.avx512.mask.pmaxu.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask)
+  %res1 = call <4 x i64> @llvm.x86.avx512.mask.pmaxu.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i64> %res, %res1
+  ret <4 x i64> %res2
+}
+
+declare <4 x i32> @llvm.x86.avx512.mask.pmins.d.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_d_128
+; CHECK-NOT: call 
+; CHECK: vpminsd %xmm
+; CHECK: {%k1} 
+define <4 x i32>@test_int_x86_avx512_mask_pmins_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %mask) {
+  %res = call <4 x i32> @llvm.x86.avx512.mask.pmins.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %mask)
+  %res1 = call <4 x i32> @llvm.x86.avx512.mask.pmins.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i32> %res, %res1
+  ret <4 x i32> %res2
+}
+
+declare <8 x i32> @llvm.x86.avx512.mask.pmins.d.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_d_256
+; CHECK-NOT: call 
+; CHECK: vpminsd %ymm
+; CHECK: {%k1} 
+define <8 x i32>@test_int_x86_avx512_mask_pmins_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3) {
+  %res = call <8 x i32> @llvm.x86.avx512.mask.pmins.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
+  %res1 = call <8 x i32> @llvm.x86.avx512.mask.pmins.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
+  %res2 = add <8 x i32> %res, %res1
+  ret <8 x i32> %res2
+}
+
+declare <2 x i64> @llvm.x86.avx512.mask.pmins.q.128(<2 x i64>, <2 x i64>, <2 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_q_128
+; CHECK-NOT: call 
+; CHECK: vpminsq %xmm
+; CHECK: {%k1} 
+define <2 x i64>@test_int_x86_avx512_mask_pmins_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
+  %res = call <2 x i64> @llvm.x86.avx512.mask.pmins.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3)
+  %res1 = call <2 x i64> @llvm.x86.avx512.mask.pmins.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 -1)
+  %res2 = add <2 x i64> %res, %res1
+  ret <2 x i64> %res2
+}
+
+declare <4 x i64> @llvm.x86.avx512.mask.pmins.q.256(<4 x i64>, <4 x i64>, <4 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pmins_q_256
+; CHECK-NOT: call 
+; CHECK: vpminsq %ymm
+; CHECK: {%k1} 
+define <4 x i64>@test_int_x86_avx512_mask_pmins_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask) {
+  %res = call <4 x i64> @llvm.x86.avx512.mask.pmins.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask)
+  %res1 = call <4 x i64> @llvm.x86.avx512.mask.pmins.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i64> %res, %res1
+  ret <4 x i64> %res2
+}
+
+declare <4 x i32> @llvm.x86.avx512.mask.pminu.d.128(<4 x i32>, <4 x i32>, <4 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_d_128
+; CHECK-NOT: call 
+; CHECK: vpminud %xmm
+; CHECK: {%k1} 
+define <4 x i32>@test_int_x86_avx512_mask_pminu_d_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %mask) {
+  %res = call <4 x i32> @llvm.x86.avx512.mask.pminu.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, i8 %mask)
+  %res1 = call <4 x i32> @llvm.x86.avx512.mask.pminu.d.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i32> %res, %res1
+  ret <4 x i32> %res2
+}
+
+declare <8 x i32> @llvm.x86.avx512.mask.pminu.d.256(<8 x i32>, <8 x i32>, <8 x i32>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_d_256
+; CHECK-NOT: call 
+; CHECK: vpminud %ymm
+; CHECK: {%k1} 
+define <8 x i32>@test_int_x86_avx512_mask_pminu_d_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3) {
+  %res = call <8 x i32> @llvm.x86.avx512.mask.pminu.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 %x3)
+  %res1 = call <8 x i32> @llvm.x86.avx512.mask.pminu.d.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, i8 -1)
+  %res2 = add <8 x i32> %res, %res1
+  ret <8 x i32> %res2
+}
+
+declare <2 x i64> @llvm.x86.avx512.mask.pminu.q.128(<2 x i64>, <2 x i64>, <2 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_q_128
+; CHECK-NOT: call 
+; CHECK: vpminuq %xmm
+; CHECK: {%k1} 
+define <2 x i64>@test_int_x86_avx512_mask_pminu_q_128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3) {
+  %res = call <2 x i64> @llvm.x86.avx512.mask.pminu.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 %x3)
+  %res1 = call <2 x i64> @llvm.x86.avx512.mask.pminu.q.128(<2 x i64> %x0, <2 x i64> %x1, <2 x i64> %x2, i8 -1)
+  %res2 = add <2 x i64> %res, %res1
+  ret <2 x i64> %res2
+}
+
+declare <4 x i64> @llvm.x86.avx512.mask.pminu.q.256(<4 x i64>, <4 x i64>, <4 x i64>, i8)
+
+; CHECK-LABEL: @test_int_x86_avx512_mask_pminu_q_256
+; CHECK-NOT: call 
+; CHECK: vpminuq %ymm
+; CHECK: {%k1} 
+define <4 x i64>@test_int_x86_avx512_mask_pminu_q_256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask) {
+  %res = call <4 x i64> @llvm.x86.avx512.mask.pminu.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> %x2, i8 %mask)
+  %res1 = call <4 x i64> @llvm.x86.avx512.mask.pminu.q.256(<4 x i64> %x0, <4 x i64> %x1, <4 x i64> zeroinitializer, i8 %mask)
+  %res2 = add <4 x i64> %res, %res1
+  ret <4 x i64> %res2
+}