[X86][SSE] Auto upgrade PADDS/PSUBS intrinsics to SADD_SAT/SSUB_SAT generic intrinsics (llvm)

[RKSimon]

Summary

This auto upgrades the signed SSE saturated math intrinsics to SADD_SAT/SSUB_SAT generic intrinsics.

There are a few items to note:

msan_x86intrinsics.ll - I changed this to another similar x86 intrinsic instead of replacing it with the generic @llvm.sadd.sat.v8i16

x86-adds-subs.ll - Not sure what's best here, the diff is that the generic intrinsics don't constant fold undefs - do we want to add support for that? Should I move this out of the x86 folder and use generic intrinsics?

pic-load-remat.ll - This was using the old sse2 intrinsics, and now use the generics. I don't think it matters but I can change it to another x86 intrinsic if that helps. The test is very old, I'm not sure if its very useful tbh.

Clang counterpart: https://reviews.llvm.org/D55890

[craig.topper]

test/CodeGen/X86/pic-load-remat.ll

1

Looks like its only looking for a load on psllw so it probably doesn't matter what else is going on. I'm fine with the generic intrinsics.

test/Transforms/InstCombine/X86/x86-adds-subs.ll

22

Probably should move this out to generic. Feels weird to test upgraded intrinsics.

[RKSimon]

test/Transforms/InstCombine/X86/x86-adds-subs.ll

22

OK - will do this as a pre-commit if that's OK? What about the UNDEF handling, just raise a bug?

[craig.topper]

test/Transforms/InstCombine/X86/x86-adds-subs.ll

22

Thats fine with me. Raise a bug for undef handling. It look like for scalars we fold undef input to -1 if I'm reading the code in InstSimplify correctly. That looks to be different than what we were previously doing for the intrinsics.

[RKSimon]

rebase

[RKSimon]

rebased

msan_x86intrinsics.ll - committed at rL349739

x86-adds-subs.ll - moved at rL349759 and raised undef regression at PR40110

pic-load-remat.ll - comitted at rL349742

[craig.topper]

LGTM

include/llvm/IR/IntrinsicsX86.td

def int_x86_sse2_ucomige_sd : GCCBuiltin<"__builtin_ia32_ucomisdge">,
                          llvm_v2f64_ty], [IntrNoMem]>;
   def int_x86_sse2_ucomineq_sd : GCCBuiltin<"__builtin_ia32_ucomisdneq">,
               Intrinsic<[llvm_i32_ty], [llvm_v2f64_ty,
                          llvm_v2f64_ty], [IntrNoMem]>;
 }
 
 // Integer arithmetic ops.
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
-  def int_x86_sse2_padds_b : GCCBuiltin<"__builtin_ia32_paddsb128">,
-              Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty,
-                         llvm_v16i8_ty], [IntrNoMem, Commutative]>;
-  def int_x86_sse2_padds_w : GCCBuiltin<"__builtin_ia32_paddsw128">,
-              Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty,
-                         llvm_v8i16_ty], [IntrNoMem, Commutative]>;
-  def int_x86_sse2_psubs_b : GCCBuiltin<"__builtin_ia32_psubsb128">,
-              Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty,
-                         llvm_v16i8_ty], [IntrNoMem]>;
-  def int_x86_sse2_psubs_w : GCCBuiltin<"__builtin_ia32_psubsw128">,
-              Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty,
-                         llvm_v8i16_ty], [IntrNoMem]>;
   def int_x86_sse2_pmulhu_w : GCCBuiltin<"__builtin_ia32_pmulhuw128">,
               Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty,
                          llvm_v8i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_sse2_pmulh_w : GCCBuiltin<"__builtin_ia32_pmulhw128">,
               Intrinsic<[llvm_v8i16_ty], [llvm_v8i16_ty,
                          llvm_v8i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_sse2_pmadd_wd : GCCBuiltin<"__builtin_ia32_pmaddwd128">,
               Intrinsic<[llvm_v4i32_ty], [llvm_v8i16_ty,
Intrinsic<[llvm_i64_ty],
               [IntrNoMem]>;
 }
 
 //===----------------------------------------------------------------------===//
 // AVX2
 
 // Integer arithmetic ops.
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
-  def int_x86_avx2_padds_b : GCCBuiltin<"__builtin_ia32_paddsb256">,
-              Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty,
-                         llvm_v32i8_ty], [IntrNoMem, Commutative]>;
-  def int_x86_avx2_padds_w : GCCBuiltin<"__builtin_ia32_paddsw256">,
-              Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty,
-                         llvm_v16i16_ty], [IntrNoMem, Commutative]>;
-  def int_x86_avx2_psubs_b : GCCBuiltin<"__builtin_ia32_psubsb256">,
-              Intrinsic<[llvm_v32i8_ty], [llvm_v32i8_ty,
-                         llvm_v32i8_ty], [IntrNoMem]>;
-  def int_x86_avx2_psubs_w : GCCBuiltin<"__builtin_ia32_psubsw256">,
-              Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty,
-                         llvm_v16i16_ty], [IntrNoMem]>;
   def int_x86_avx2_pmulhu_w : GCCBuiltin<"__builtin_ia32_pmulhuw256">,
               Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty,
                          llvm_v16i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx2_pmulh_w : GCCBuiltin<"__builtin_ia32_pmulhw256">,
               Intrinsic<[llvm_v16i16_ty], [llvm_v16i16_ty,
                          llvm_v16i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx2_pmadd_wd : GCCBuiltin<"__builtin_ia32_pmaddwd256">,
               Intrinsic<[llvm_v8i32_ty], [llvm_v16i16_ty,
def int_x86_avx512_rsqrt28_sd : GCCBuiltin<"__builtin_ia32_rsqrt28sd_round_mask">,
                                         llvm_v2f64_ty, llvm_i8_ty, llvm_i32_ty],
                       [IntrNoMem]>;
   def int_x86_avx512_psad_bw_512 : GCCBuiltin<"__builtin_ia32_psadbw512">,
               Intrinsic<[llvm_v8i64_ty], [llvm_v64i8_ty, llvm_v64i8_ty],
                         [IntrNoMem, Commutative]>;
 }
 // Integer arithmetic ops
 let TargetPrefix = "x86" in {
-  def int_x86_avx512_padds_b_512 : GCCBuiltin<"__builtin_ia32_paddsb512">,
-          Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty],
-                    [IntrNoMem]>;
-  def int_x86_avx512_padds_w_512 : GCCBuiltin<"__builtin_ia32_paddsw512">,
-          Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty],
-                     [IntrNoMem]>;
-  def int_x86_avx512_psubs_b_512 : GCCBuiltin<"__builtin_ia32_psubsb512">,
-          Intrinsic<[llvm_v64i8_ty], [llvm_v64i8_ty, llvm_v64i8_ty],
-                     [IntrNoMem]>;
-  def int_x86_avx512_psubs_w_512 : GCCBuiltin<"__builtin_ia32_psubsw512">,
-          Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty, llvm_v32i16_ty],
-                     [IntrNoMem]>;
   def int_x86_avx512_pmulhu_w_512 : GCCBuiltin<"__builtin_ia32_pmulhuw512">,
               Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty,
                          llvm_v32i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx512_pmulh_w_512 : GCCBuiltin<"__builtin_ia32_pmulhw512">,
               Intrinsic<[llvm_v32i16_ty], [llvm_v32i16_ty,
                          llvm_v32i16_ty], [IntrNoMem, Commutative]>;
   def int_x86_avx512_pmaddw_d_512 : GCCBuiltin<"__builtin_ia32_pmaddwd512">,
               Intrinsic<[llvm_v16i32_ty], [llvm_v32i16_ty,

lib/IR/AutoUpgrade.cpp

static bool ShouldUpgradeX86Intrinsic(Function *F, StringRef Name) {
   // intrinsics. It is currently undecided how we will determine that future
   // point.
   if (Name == "addcarryx.u32" || // Added in 8.0
       Name == "addcarryx.u64" || // Added in 8.0
       Name == "addcarry.u32" || // Added in 8.0
       Name == "addcarry.u64" || // Added in 8.0
       Name == "subborrow.u32" || // Added in 8.0
       Name == "subborrow.u64" || // Added in 8.0
+      Name.startswith("sse2.padds.") || // Added in 8.0
+      Name.startswith("sse2.psubs.") || // Added in 8.0
       Name.startswith("sse2.paddus.") || // Added in 8.0
       Name.startswith("sse2.psubus.") || // Added in 8.0
+      Name.startswith("avx2.padds.") || // Added in 8.0
+      Name.startswith("avx2.psubs.") || // Added in 8.0
       Name.startswith("avx2.paddus.") || // Added in 8.0
       Name.startswith("avx2.psubus.") || // Added in 8.0
+      Name.startswith("avx512.padds.") || // Added in 8.0
+      Name.startswith("avx512.psubs.") || // Added in 8.0
+      Name.startswith("avx512.mask.padds.") || // Added in 8.0
+      Name.startswith("avx512.mask.psubs.") || // Added in 8.0
       Name.startswith("avx512.mask.paddus.") || // Added in 8.0
       Name.startswith("avx512.mask.psubus.") || // Added in 8.0
       Name=="ssse3.pabs.b.128" || // Added in 6.0
       Name=="ssse3.pabs.w.128" || // Added in 6.0
       Name=="ssse3.pabs.d.128" || // Added in 6.0
       Name.startswith("fma4.vfmadd.s") || // Added in 7.0
       Name.startswith("fma.vfmadd.") || // Added in 7.0
       Name.startswith("fma.vfmsub.") || // Added in 7.0
if (Name == "addcarryx.u32" || // Added in 8.0
       Name.startswith("avx512.mask.div.p") || // Added in 7.0. 128/256 in 4.0
       Name.startswith("avx512.mask.max.p") || // Added in 7.0. 128/256 in 5.0
       Name.startswith("avx512.mask.min.p") || // Added in 7.0. 128/256 in 5.0
       Name.startswith("avx512.mask.fpclass.p") || // Added in 7.0
       Name.startswith("avx512.mask.prorv.") || // Added in 7.0
       Name.startswith("avx512.mask.pror.") || // Added in 7.0
       Name.startswith("avx512.mask.prolv.") || // Added in 7.0
       Name.startswith("avx512.mask.prol.") || // Added in 7.0
-      Name.startswith("avx512.mask.padds.") || // Added in 8.0
-      Name.startswith("avx512.mask.psubs.") || // Added in 8.0
       Name == "sse.cvtsi2ss" || // Added in 7.0
       Name == "sse.cvtsi642ss" || // Added in 7.0
       Name == "sse2.cvtsi2sd" || // Added in 7.0
       Name == "sse2.cvtsi642sd" || // Added in 7.0
       Name == "sse2.cvtss2sd" || // Added in 7.0
       Name == "sse2.cvtdq2pd" || // Added in 3.9
       Name == "sse2.cvtdq2ps" || // Added in 7.0
       Name == "sse2.cvtps2pd" || // Added in 3.9
static Value *UpgradeX86ALIGNIntrinsics(IRBuilder<> &Builder, Value *Op0,
   Value *Align = Builder.CreateShuffleVector(Op1, Op0,
                                              makeArrayRef(Indices, NumElts),
                                              "palignr");
 
   return EmitX86Select(Builder, Mask, Align, Passthru);
 }
 
 static Value *UpgradeX86AddSubSatIntrinsics(IRBuilder<> &Builder, CallInst &CI,
-                                            bool IsAddition) {
+                                            bool IsSigned, bool IsAddition) {
   Type *Ty = CI.getType();
   Value *Op0 = CI.getOperand(0);
   Value *Op1 = CI.getOperand(1);
 
-  Intrinsic::ID IID = IsAddition ? Intrinsic::uadd_sat : Intrinsic::usub_sat;
+  Intrinsic::ID IID =
+      IsSigned ? (IsAddition ? Intrinsic::sadd_sat : Intrinsic::ssub_sat)
+               : (IsAddition ? Intrinsic::uadd_sat : Intrinsic::usub_sat);
   Function *Intrin = Intrinsic::getDeclaration(CI.getModule(), IID, Ty);
   Value *Res = Builder.CreateCall(Intrin, {Op0, Op1});
 
   if (CI.getNumArgOperands() == 4) { // For masked intrinsics.
     Value *VecSrc = CI.getOperand(2);
     Value *Mask = CI.getOperand(3);
     Res = EmitX86Select(Builder, Mask, Res, VecSrc);
   }
if (Name.startswith("max.p")) {
     else if (VecWidth == 128 && EltWidth == 64)
       IID = Intrinsic::x86_avx512_prol_q_128;
     else if (VecWidth == 256 && EltWidth == 64)
       IID = Intrinsic::x86_avx512_prol_q_256;
     else if (VecWidth == 512 && EltWidth == 64)
       IID = Intrinsic::x86_avx512_prol_q_512;
     else
       llvm_unreachable("Unexpected intrinsic");
-  } else if (Name.startswith("padds.")) {
-    if (VecWidth == 128 && EltWidth == 8)
-      IID = Intrinsic::x86_sse2_padds_b;
-    else if (VecWidth == 256 && EltWidth == 8)
-      IID = Intrinsic::x86_avx2_padds_b;
-    else if (VecWidth == 512 && EltWidth == 8)
-      IID = Intrinsic::x86_avx512_padds_b_512;
-    else if (VecWidth == 128 && EltWidth == 16)
-      IID = Intrinsic::x86_sse2_padds_w;
-    else if (VecWidth == 256 && EltWidth == 16)
-      IID = Intrinsic::x86_avx2_padds_w;
-    else if (VecWidth == 512 && EltWidth == 16)
-      IID = Intrinsic::x86_avx512_padds_w_512;
-    else
-      llvm_unreachable("Unexpected intrinsic");
-  } else if (Name.startswith("psubs.")) {
-    if (VecWidth == 128 && EltWidth == 8)
-      IID = Intrinsic::x86_sse2_psubs_b;
-    else if (VecWidth == 256 && EltWidth == 8)
-      IID = Intrinsic::x86_avx2_psubs_b;
-    else if (VecWidth == 512 && EltWidth == 8)
-      IID = Intrinsic::x86_avx512_psubs_b_512;
-    else if (VecWidth == 128 && EltWidth == 16)
-      IID = Intrinsic::x86_sse2_psubs_w;
-    else if (VecWidth == 256 && EltWidth == 16)
-      IID = Intrinsic::x86_avx2_psubs_w;
-    else if (VecWidth == 512 && EltWidth == 16)
-      IID = Intrinsic::x86_avx512_psubs_w_512;
-    else
-      llvm_unreachable("Unexpected intrinsic");
   } else
     return false;
 
   SmallVector<Value *, 4> Args(CI.arg_operands().begin(),
                                CI.arg_operands().end());
   Args.pop_back();
   Args.pop_back();
   Rep = Builder.CreateCall(Intrinsic::getDeclaration(CI.getModule(), IID),
if (IsX86 && (Name.startswith("sse2.pcmp") ||
       unsigned NumElts = CI->getType()->getVectorNumElements();
       Type *MaskTy = VectorType::get(Type::getInt32Ty(C), NumElts);
       Rep = Builder.CreateShuffleVector(Op, UndefValue::get(Op->getType()),
                                         Constant::getNullValue(MaskTy));
 
       if (CI->getNumArgOperands() == 3)
         Rep = EmitX86Select(Builder, CI->getArgOperand(2), Rep,
                             CI->getArgOperand(1));
+    } else if (IsX86 && (Name.startswith("sse2.padds.") ||
+                         Name.startswith("sse2.psubs.") ||
+                         Name.startswith("avx2.padds.") ||
+                         Name.startswith("avx2.psubs.") ||
+                         Name.startswith("avx512.padds.") ||
+                         Name.startswith("avx512.psubs.") ||
+                         Name.startswith("avx512.mask.padds.") ||
+                         Name.startswith("avx512.mask.psubs."))) {
+      bool IsAdd = Name.contains(".padds");
+      Rep = UpgradeX86AddSubSatIntrinsics(Builder, *CI, true, IsAdd);
     } else if (IsX86 && (Name.startswith("sse2.paddus.") ||
                          Name.startswith("sse2.psubus.") ||
                          Name.startswith("avx2.paddus.") ||
                          Name.startswith("avx2.psubus.") ||
                          Name.startswith("avx512.mask.paddus.") ||
                          Name.startswith("avx512.mask.psubus."))) {
       bool IsAdd = Name.contains(".paddus");
-      Rep = UpgradeX86AddSubSatIntrinsics(Builder, *CI, IsAdd);
+      Rep = UpgradeX86AddSubSatIntrinsics(Builder, *CI, false, IsAdd);
     } else if (IsX86 && Name.startswith("avx512.mask.palignr.")) {
       Rep = UpgradeX86ALIGNIntrinsics(Builder, CI->getArgOperand(0),
                                       CI->getArgOperand(1),
                                       CI->getArgOperand(2),
                                       CI->getArgOperand(3),
                                       CI->getArgOperand(4),
                                       false);
     } else if (IsX86 && Name.startswith("avx512.mask.valign.")) {

lib/Target/X86/X86IntrinsicsInfo.h

static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx_vpermilvar_pd,     INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_pd_256, INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_ps,     INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx_vpermilvar_ps_256, INTR_TYPE_2OP, X86ISD::VPERMILPV, 0),
   X86_INTRINSIC_DATA(avx2_packssdw, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx2_packsswb, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx2_packusdw, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
   X86_INTRINSIC_DATA(avx2_packuswb, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
-  X86_INTRINSIC_DATA(avx2_padds_b, INTR_TYPE_2OP, ISD::SADDSAT, 0),
-  X86_INTRINSIC_DATA(avx2_padds_w, INTR_TYPE_2OP, ISD::SADDSAT, 0),
   X86_INTRINSIC_DATA(avx2_permd, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx2_permps, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx2_phadd_d, INTR_TYPE_2OP, X86ISD::HADD, 0),
   X86_INTRINSIC_DATA(avx2_phadd_w, INTR_TYPE_2OP, X86ISD::HADD, 0),
   X86_INTRINSIC_DATA(avx2_phsub_d, INTR_TYPE_2OP, X86ISD::HSUB, 0),
   X86_INTRINSIC_DATA(avx2_phsub_w, INTR_TYPE_2OP, X86ISD::HSUB, 0),
   X86_INTRINSIC_DATA(avx2_pmadd_ub_sw, INTR_TYPE_2OP, X86ISD::VPMADDUBSW, 0),
   X86_INTRINSIC_DATA(avx2_pmadd_wd, INTR_TYPE_2OP, X86ISD::VPMADDWD, 0),
static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx2_psrl_w, INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(avx2_psrli_d, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx2_psrli_q, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx2_psrli_w, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx2_psrlv_d, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx2_psrlv_d_256, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx2_psrlv_q, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx2_psrlv_q_256, INTR_TYPE_2OP, ISD::SRL, 0),
-  X86_INTRINSIC_DATA(avx2_psubs_b, INTR_TYPE_2OP, ISD::SSUBSAT, 0),
-  X86_INTRINSIC_DATA(avx2_psubs_w, INTR_TYPE_2OP, ISD::SSUBSAT, 0),
   X86_INTRINSIC_DATA(avx512_add_pd_512, INTR_TYPE_2OP, ISD::FADD, X86ISD::FADD_RND),
   X86_INTRINSIC_DATA(avx512_add_ps_512, INTR_TYPE_2OP, ISD::FADD, X86ISD::FADD_RND),
   X86_INTRINSIC_DATA(avx512_cmp_pd_128, CMP_MASK_CC, X86ISD::CMPM, 0),
   X86_INTRINSIC_DATA(avx512_cmp_pd_256, CMP_MASK_CC, X86ISD::CMPM, 0),
   X86_INTRINSIC_DATA(avx512_cmp_pd_512, CMP_MASK_CC, X86ISD::CMPM, X86ISD::CMPM_RND),
   X86_INTRINSIC_DATA(avx512_cmp_ps_128, CMP_MASK_CC, X86ISD::CMPM, 0),
   X86_INTRINSIC_DATA(avx512_cmp_ps_256, CMP_MASK_CC, X86ISD::CMPM, 0),
   X86_INTRINSIC_DATA(avx512_cmp_ps_512, CMP_MASK_CC, X86ISD::CMPM, X86ISD::CMPM_RND),
static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_min_pd_512, INTR_TYPE_2OP, X86ISD::FMIN, X86ISD::FMIN_RND),
   X86_INTRINSIC_DATA(avx512_min_ps_512, INTR_TYPE_2OP, X86ISD::FMIN, X86ISD::FMIN_RND),
   X86_INTRINSIC_DATA(avx512_mul_pd_512, INTR_TYPE_2OP, ISD::FMUL, X86ISD::FMUL_RND),
   X86_INTRINSIC_DATA(avx512_mul_ps_512, INTR_TYPE_2OP, ISD::FMUL, X86ISD::FMUL_RND),
   X86_INTRINSIC_DATA(avx512_packssdw_512, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_packsswb_512, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(avx512_packusdw_512, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
   X86_INTRINSIC_DATA(avx512_packuswb_512, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
-  X86_INTRINSIC_DATA(avx512_padds_b_512, INTR_TYPE_2OP, ISD::SADDSAT, 0),
-  X86_INTRINSIC_DATA(avx512_padds_w_512, INTR_TYPE_2OP, ISD::SADDSAT, 0),
   X86_INTRINSIC_DATA(avx512_permvar_df_256, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_df_512, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_di_256, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_di_512, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_hi_128, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_hi_256, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_hi_512, VPERM_2OP, X86ISD::VPERMV, 0),
   X86_INTRINSIC_DATA(avx512_permvar_qi_128, VPERM_2OP, X86ISD::VPERMV, 0),
static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_psrli_d_512, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx512_psrli_q_512, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx512_psrli_w_512, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx512_psrlv_d_512, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx512_psrlv_q_512, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx512_psrlv_w_128, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx512_psrlv_w_256, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx512_psrlv_w_512, INTR_TYPE_2OP, ISD::SRL, 0),
-  X86_INTRINSIC_DATA(avx512_psubs_b_512, INTR_TYPE_2OP, ISD::SSUBSAT, 0),
-  X86_INTRINSIC_DATA(avx512_psubs_w_512, INTR_TYPE_2OP, ISD::SSUBSAT, 0),
   X86_INTRINSIC_DATA(avx512_pternlog_d_128, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
   X86_INTRINSIC_DATA(avx512_pternlog_d_256, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
   X86_INTRINSIC_DATA(avx512_pternlog_d_512, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
   X86_INTRINSIC_DATA(avx512_pternlog_q_128, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
   X86_INTRINSIC_DATA(avx512_pternlog_q_256, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
   X86_INTRINSIC_DATA(avx512_pternlog_q_512, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_pd_128, INTR_TYPE_1OP_MASK, X86ISD::RCP14, 0),
   X86_INTRINSIC_DATA(avx512_rcp14_pd_256, INTR_TYPE_1OP_MASK, X86ISD::RCP14, 0),
static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(sse2_max_pd,       INTR_TYPE_2OP, X86ISD::FMAX, 0),
   X86_INTRINSIC_DATA(sse2_max_sd,       INTR_TYPE_2OP, X86ISD::FMAXS, 0),
   X86_INTRINSIC_DATA(sse2_min_pd,       INTR_TYPE_2OP, X86ISD::FMIN, 0),
   X86_INTRINSIC_DATA(sse2_min_sd,       INTR_TYPE_2OP, X86ISD::FMINS, 0),
   X86_INTRINSIC_DATA(sse2_movmsk_pd,    INTR_TYPE_1OP, X86ISD::MOVMSK, 0),
   X86_INTRINSIC_DATA(sse2_packssdw_128, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(sse2_packsswb_128, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
   X86_INTRINSIC_DATA(sse2_packuswb_128, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
-  X86_INTRINSIC_DATA(sse2_padds_b,      INTR_TYPE_2OP, ISD::SADDSAT, 0),
-  X86_INTRINSIC_DATA(sse2_padds_w,      INTR_TYPE_2OP, ISD::SADDSAT, 0),
   X86_INTRINSIC_DATA(sse2_pmadd_wd,     INTR_TYPE_2OP, X86ISD::VPMADDWD, 0),
   X86_INTRINSIC_DATA(sse2_pmovmskb_128, INTR_TYPE_1OP, X86ISD::MOVMSK, 0),
   X86_INTRINSIC_DATA(sse2_pmulh_w,      INTR_TYPE_2OP, ISD::MULHS, 0),
   X86_INTRINSIC_DATA(sse2_pmulhu_w,     INTR_TYPE_2OP, ISD::MULHU, 0),
   X86_INTRINSIC_DATA(sse2_psad_bw,      INTR_TYPE_2OP, X86ISD::PSADBW, 0),
   X86_INTRINSIC_DATA(sse2_psll_d,       INTR_TYPE_2OP, X86ISD::VSHL, 0),
   X86_INTRINSIC_DATA(sse2_psll_q,       INTR_TYPE_2OP, X86ISD::VSHL, 0),
   X86_INTRINSIC_DATA(sse2_psll_w,       INTR_TYPE_2OP, X86ISD::VSHL, 0),
   X86_INTRINSIC_DATA(sse2_pslli_d,      VSHIFT, X86ISD::VSHLI, 0),
   X86_INTRINSIC_DATA(sse2_pslli_q,      VSHIFT, X86ISD::VSHLI, 0),
   X86_INTRINSIC_DATA(sse2_pslli_w,      VSHIFT, X86ISD::VSHLI, 0),
   X86_INTRINSIC_DATA(sse2_psra_d,       INTR_TYPE_2OP, X86ISD::VSRA, 0),
   X86_INTRINSIC_DATA(sse2_psra_w,       INTR_TYPE_2OP, X86ISD::VSRA, 0),
   X86_INTRINSIC_DATA(sse2_psrai_d,      VSHIFT, X86ISD::VSRAI, 0),
   X86_INTRINSIC_DATA(sse2_psrai_w,      VSHIFT, X86ISD::VSRAI, 0),
   X86_INTRINSIC_DATA(sse2_psrl_d,       INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(sse2_psrl_q,       INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(sse2_psrl_w,       INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(sse2_psrli_d,      VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(sse2_psrli_q,      VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(sse2_psrli_w,      VSHIFT, X86ISD::VSRLI, 0),
-  X86_INTRINSIC_DATA(sse2_psubs_b,      INTR_TYPE_2OP, ISD::SSUBSAT, 0),
-  X86_INTRINSIC_DATA(sse2_psubs_w,      INTR_TYPE_2OP, ISD::SSUBSAT, 0),
   X86_INTRINSIC_DATA(sse2_ucomieq_sd,   COMI, X86ISD::UCOMI, ISD::SETEQ),
   X86_INTRINSIC_DATA(sse2_ucomige_sd,   COMI, X86ISD::UCOMI, ISD::SETGE),
   X86_INTRINSIC_DATA(sse2_ucomigt_sd,   COMI, X86ISD::UCOMI, ISD::SETGT),
   X86_INTRINSIC_DATA(sse2_ucomile_sd,   COMI, X86ISD::UCOMI, ISD::SETLE),
   X86_INTRINSIC_DATA(sse2_ucomilt_sd,   COMI, X86ISD::UCOMI, ISD::SETLT),
   X86_INTRINSIC_DATA(sse2_ucomineq_sd,  COMI, X86ISD::UCOMI, ISD::SETNE),
   X86_INTRINSIC_DATA(sse3_addsub_pd,    INTR_TYPE_2OP, X86ISD::ADDSUB, 0),
   X86_INTRINSIC_DATA(sse3_addsub_ps,    INTR_TYPE_2OP, X86ISD::ADDSUB, 0),

lib/Transforms/InstCombine/InstCombineCalls.cpp

if (Len <= 8 && isPowerOf2_32((uint32_t)Len)) {
     // Set the size of the copy to 0, it will be deleted on the next iteration.
     MI->setLength(Constant::getNullValue(LenC->getType()));
     return MI;
   }
 
   return nullptr;
 }
 
-static Value *simplifyX86AddsSubs(const IntrinsicInst &II,
-                                  InstCombiner::BuilderTy &Builder) {
-  bool IsAddition;
-
-  switch (II.getIntrinsicID()) {
-  default: llvm_unreachable("Unexpected intrinsic!");
-  case Intrinsic::x86_sse2_padds_b:
-  case Intrinsic::x86_sse2_padds_w:
-  case Intrinsic::x86_avx2_padds_b:
-  case Intrinsic::x86_avx2_padds_w:
-  case Intrinsic::x86_avx512_padds_b_512:
-  case Intrinsic::x86_avx512_padds_w_512:
-    IsAddition = true;
-    break;
-  case Intrinsic::x86_sse2_psubs_b:
-  case Intrinsic::x86_sse2_psubs_w:
-  case Intrinsic::x86_avx2_psubs_b:
-  case Intrinsic::x86_avx2_psubs_w:
-  case Intrinsic::x86_avx512_psubs_b_512:
-  case Intrinsic::x86_avx512_psubs_w_512:
-    IsAddition = false;
-    break;
-  }
-
-  auto *Arg0 = dyn_cast<Constant>(II.getOperand(0));
-  auto *Arg1 = dyn_cast<Constant>(II.getOperand(1));
-  auto VT = cast<VectorType>(II.getType());
-  auto SVT = VT->getElementType();
-  unsigned NumElems = VT->getNumElements();
-
-  if (!Arg0 || !Arg1)
-    return nullptr;
-
-  SmallVector<Constant *, 64> Result;
-
-  APInt MaxValue = APInt::getSignedMaxValue(SVT->getIntegerBitWidth());
-  APInt MinValue = APInt::getSignedMinValue(SVT->getIntegerBitWidth());
-  for (unsigned i = 0; i < NumElems; ++i) {
-    auto *Elt0 = Arg0->getAggregateElement(i);
-    auto *Elt1 = Arg1->getAggregateElement(i);
-    if (isa<UndefValue>(Elt0) || isa<UndefValue>(Elt1)) {
-      Result.push_back(UndefValue::get(SVT));
-      continue;
-    }
-
-    if (!isa<ConstantInt>(Elt0) || !isa<ConstantInt>(Elt1))
-      return nullptr;
-
-    const APInt &Val0 = cast<ConstantInt>(Elt0)->getValue();
-    const APInt &Val1 = cast<ConstantInt>(Elt1)->getValue();
-    bool Overflow = false;
-    APInt ResultElem = IsAddition ? Val0.sadd_ov(Val1, Overflow)
-                                  : Val0.ssub_ov(Val1, Overflow);
-    if (Overflow)
-      ResultElem = Val0.isNegative() ? MinValue : MaxValue;
-    Result.push_back(Constant::getIntegerValue(SVT, ResultElem));
-  }
-
-  return ConstantVector::get(Result);
-}
-
 static Value *simplifyX86immShift(const IntrinsicInst &II,
                                   InstCombiner::BuilderTy &Builder) {
   bool LogicalShift = false;
   bool ShiftLeft = false;
 
   switch (II.getIntrinsicID()) {
   default: llvm_unreachable("Unexpected intrinsic!");
   case Intrinsic::x86_sse2_psra_d:
if (Value *V = SimplifyDemandedVectorElts(II, AllOnesEltMask, UndefElts)) {
       if (V != II)
         return replaceInstUsesWith(*II, V);
       return II;
     } else if (Value *V = simplifyX86round(*II, Builder))
       return replaceInstUsesWith(*II, V);
     break;
   }
 
-  // Constant fold add/sub with saturation intrinsics.
-  case Intrinsic::x86_sse2_padds_b:
-  case Intrinsic::x86_sse2_padds_w:
-  case Intrinsic::x86_sse2_psubs_b:
-  case Intrinsic::x86_sse2_psubs_w:
-  case Intrinsic::x86_avx2_padds_b:
-  case Intrinsic::x86_avx2_padds_w:
-  case Intrinsic::x86_avx2_psubs_b:
-  case Intrinsic::x86_avx2_psubs_w:
-  case Intrinsic::x86_avx512_padds_b_512:
-  case Intrinsic::x86_avx512_padds_w_512:
-  case Intrinsic::x86_avx512_psubs_b_512:
-  case Intrinsic::x86_avx512_psubs_w_512:
-    if (Value *V = simplifyX86AddsSubs(*II, Builder))
-      return replaceInstUsesWith(*II, V);
-    break;
-
   // Constant fold ashr( <A x Bi>, Ci ).
   // Constant fold lshr( <A x Bi>, Ci ).
   // Constant fold shl( <A x Bi>, Ci ).
   case Intrinsic::x86_sse2_psrai_d:
   case Intrinsic::x86_sse2_psrai_w:
   case Intrinsic::x86_avx2_psrai_d:
   case Intrinsic::x86_avx2_psrai_w:
   case Intrinsic::x86_avx512_psrai_q_128:

test/CodeGen/X86/avx2-intrinsics-fast-isel.ll

 
 define <4 x i64> @test_mm256_adds_epi8(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_mm256_adds_epi8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %arg0 = bitcast <4 x i64> %a0 to <32 x i8>
   %arg1 = bitcast <4 x i64> %a1 to <32 x i8>
-  %res = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %arg0, <32 x i8> %arg1)
+  %res = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %arg0, <32 x i8> %arg1)
   %bc = bitcast <32 x i8> %res to <4 x i64>
   ret <4 x i64> %bc
 }
-declare <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8>, <32 x i8>) nounwind readnone
+declare <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8>, <32 x i8>) nounwind readnone
 
 define <4 x i64> @test_mm256_adds_epi16(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_mm256_adds_epi16:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %arg0 = bitcast <4 x i64> %a0 to <16 x i16>
   %arg1 = bitcast <4 x i64> %a1 to <16 x i16>
-  %res = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %arg0, <16 x i16> %arg1)
+  %res = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %arg0, <16 x i16> %arg1)
   %bc = bitcast <16 x i16> %res to <4 x i64>
   ret <4 x i64> %bc
 }
-declare <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16>, <16 x i16>) nounwind readnone
+declare <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16>, <16 x i16>) nounwind readnone
 
 define <4 x i64> @test_mm256_adds_epu8(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_mm256_adds_epu8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpaddusb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %arg0 = bitcast <4 x i64> %a0 to <32 x i8>
   %arg1 = bitcast <4 x i64> %a1 to <32 x i8>

 
 define <4 x i64> @test_mm256_subs_epi8(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_mm256_subs_epi8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %arg0 = bitcast <4 x i64> %a0 to <32 x i8>
   %arg1 = bitcast <4 x i64> %a1 to <32 x i8>
-  %res = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %arg0, <32 x i8> %arg1)
+  %res = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %arg0, <32 x i8> %arg1)
   %bc = bitcast <32 x i8> %res to <4 x i64>
   ret <4 x i64> %bc
 }
-declare <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8>, <32 x i8>) nounwind readnone
+declare <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8>, <32 x i8>) nounwind readnone
 
 define <4 x i64> @test_mm256_subs_epi16(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_mm256_subs_epi16:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %arg0 = bitcast <4 x i64> %a0 to <16 x i16>
   %arg1 = bitcast <4 x i64> %a1 to <16 x i16>
-  %res = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %arg0, <16 x i16> %arg1)
+  %res = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %arg0, <16 x i16> %arg1)
   %bc = bitcast <16 x i16> %res to <4 x i64>
   ret <4 x i64> %bc
 }
-declare <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16>, <16 x i16>) nounwind readnone
+declare <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16>, <16 x i16>) nounwind readnone
 
 define <4 x i64> @test_mm256_subs_epu8(<4 x i64> %a0, <4 x i64> %a1) {
 ; CHECK-LABEL: test_mm256_subs_epu8:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsubusb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %arg0 = bitcast <4 x i64> %a0 to <32 x i8>
   %arg1 = bitcast <4 x i64> %a1 to <32 x i8>

test/CodeGen/X86/avx2-intrinsics-x86-upgrade.ll

 ; CHECK-NEXT:    vpmuldq %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res = call <4 x i64> @llvm.x86.avx2.pmul.dq(<8 x i32> %a0, <8 x i32> %a1) ; <<4 x i64>> [#uses=1]
   ret <4 x i64> %res
 }
 declare <4 x i64> @llvm.x86.avx2.pmul.dq(<8 x i32>, <8 x i32>) nounwind readnone
 
 
+define <32 x i8> @test_x86_avx2_padds_b(<32 x i8> %a0, <32 x i8> %a1) {
+; CHECK-LABEL: test_x86_avx2_padds_b:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
+  %res = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a0, <32 x i8> %a1) ; <<32 x i8>> [#uses=1]
+  ret <32 x i8> %res
+}
+declare <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8>, <32 x i8>) nounwind readnone
+
+
+define <16 x i16> @test_x86_avx2_padds_w(<16 x i16> %a0, <16 x i16> %a1) {
+; CHECK-LABEL: test_x86_avx2_padds_w:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
+  %res = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a0, <16 x i16> %a1) ; <<16 x i16>> [#uses=1]
+  ret <16 x i16> %res
+}
+declare <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16>, <16 x i16>) nounwind readnone
+
+
 define <32 x i8> @test_x86_avx2_paddus_b(<32 x i8> %a0, <32 x i8> %a1) {
 ; CHECK-LABEL: test_x86_avx2_paddus_b:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vpaddusb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res = call <32 x i8> @llvm.x86.avx2.paddus.b(<32 x i8> %a0, <32 x i8> %a1) ; <<32 x i8>> [#uses=1]
   ret <32 x i8> %res
 }
 declare <32 x i8> @llvm.x86.avx2.paddus.b(<32 x i8>, <32 x i8>) nounwind readnone
 
 
 define <16 x i16> @test_x86_avx2_paddus_w(<16 x i16> %a0, <16 x i16> %a1) {
 ; CHECK-LABEL: test_x86_avx2_paddus_w:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vpaddusw %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res = call <16 x i16> @llvm.x86.avx2.paddus.w(<16 x i16> %a0, <16 x i16> %a1) ; <<16 x i16>> [#uses=1]
   ret <16 x i16> %res
 }
 declare <16 x i16> @llvm.x86.avx2.paddus.w(<16 x i16>, <16 x i16>) nounwind readnone
 
 
+define <32 x i8> @test_x86_avx2_psubs_b(<32 x i8> %a0, <32 x i8> %a1) {
+; CHECK-LABEL: test_x86_avx2_psubs_b:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
+  %res = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a0, <32 x i8> %a1) ; <<32 x i8>> [#uses=1]
+  ret <32 x i8> %res
+}
+declare <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8>, <32 x i8>) nounwind readnone
+
+
+define <16 x i16> @test_x86_avx2_psubs_w(<16 x i16> %a0, <16 x i16> %a1) {
+; CHECK-LABEL: test_x86_avx2_psubs_w:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
+  %res = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a0, <16 x i16> %a1) ; <<16 x i16>> [#uses=1]
+  ret <16 x i16> %res
+}
+declare <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16>, <16 x i16>) nounwind readnone
+
+
 define <32 x i8> @test_x86_avx2_psubus_b(<32 x i8> %a0, <32 x i8> %a1) {
 ; CHECK-LABEL: test_x86_avx2_psubus_b:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vpsubusb %ymm1, %ymm0, %ymm0
 ; CHECK-NEXT:    ret{{[l|q]}}
   %res = call <32 x i8> @llvm.x86.avx2.psubus.b(<32 x i8> %a0, <32 x i8> %a1) ; <<32 x i8>> [#uses=1]
   ret <32 x i8> %res
 }

test/CodeGen/X86/avx2-intrinsics-x86.ll

 ; X64-AVX512VL-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x05,A,A,A,A]
 ; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI5_0-4, kind: reloc_riprel_4byte
 ; X64-AVX512VL-NEXT:    retq ## encoding: [0xc3]
   %res = call <32 x i8> @llvm.x86.avx2.packuswb(<16 x i16> <i16 0, i16 255, i16 256, i16 65535, i16 -1, i16 -255, i16 -256, i16 -32678, i16 0, i16 255, i16 256, i16 65535, i16 -1, i16 -255, i16 -256, i16 -32678>, <16 x i16> zeroinitializer)
   ret <32 x i8> %res
 }
 
 
-define <32 x i8> @test_x86_avx2_padds_b(<32 x i8> %a0, <32 x i8> %a1) {
-; AVX2-LABEL: test_x86_avx2_padds_b:
-; AVX2:       ## %bb.0:
-; AVX2-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0xec,0xc1]
-; AVX2-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512VL-LABEL: test_x86_avx2_padds_b:
-; AVX512VL:       ## %bb.0:
-; AVX512VL-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xec,0xc1]
-; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a0, <32 x i8> %a1) ; <<32 x i8>> [#uses=1]
-  ret <32 x i8> %res
-}
-declare <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8>, <32 x i8>) nounwind readnone
-
-
-define <16 x i16> @test_x86_avx2_padds_w(<16 x i16> %a0, <16 x i16> %a1) {
-; AVX2-LABEL: test_x86_avx2_padds_w:
-; AVX2:       ## %bb.0:
-; AVX2-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0xed,0xc1]
-; AVX2-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512VL-LABEL: test_x86_avx2_padds_w:
-; AVX512VL:       ## %bb.0:
-; AVX512VL-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xed,0xc1]
-; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a0, <16 x i16> %a1) ; <<16 x i16>> [#uses=1]
-  ret <16 x i16> %res
-}
-declare <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16>, <16 x i16>) nounwind readnone
-
-
 define <8 x i32> @test_x86_avx2_pmadd_wd(<16 x i16> %a0, <16 x i16> %a1) {
 ; AVX2-LABEL: test_x86_avx2_pmadd_wd:
 ; AVX2:       ## %bb.0:
 ; AVX2-NEXT:    vpmaddwd %ymm1, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0xf5,0xc1]
 ; AVX2-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
 ;
 ; AVX512VL-LABEL: test_x86_avx2_pmadd_wd:
 ; AVX512VL:       ## %bb.0:

 ; AVX512VL-NEXT:    vpsrlw $7, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x71,0xd0,0x07]
 ; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %res = call <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16> %a0, i32 7) ; <<16 x i16>> [#uses=1]
   ret <16 x i16> %res
 }
 declare <16 x i16> @llvm.x86.avx2.psrli.w(<16 x i16>, i32) nounwind readnone
 
 
-define <32 x i8> @test_x86_avx2_psubs_b(<32 x i8> %a0, <32 x i8> %a1) {
-; AVX2-LABEL: test_x86_avx2_psubs_b:
-; AVX2:       ## %bb.0:
-; AVX2-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0xe8,0xc1]
-; AVX2-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512VL-LABEL: test_x86_avx2_psubs_b:
-; AVX512VL:       ## %bb.0:
-; AVX512VL-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe8,0xc1]
-; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a0, <32 x i8> %a1) ; <<32 x i8>> [#uses=1]
-  ret <32 x i8> %res
-}
-declare <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8>, <32 x i8>) nounwind readnone
-
-
-define <16 x i16> @test_x86_avx2_psubs_w(<16 x i16> %a0, <16 x i16> %a1) {
-; AVX2-LABEL: test_x86_avx2_psubs_w:
-; AVX2:       ## %bb.0:
-; AVX2-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 ## encoding: [0xc5,0xfd,0xe9,0xc1]
-; AVX2-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512VL-LABEL: test_x86_avx2_psubs_w:
-; AVX512VL:       ## %bb.0:
-; AVX512VL-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe9,0xc1]
-; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a0, <16 x i16> %a1) ; <<16 x i16>> [#uses=1]
-  ret <16 x i16> %res
-}
-declare <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16>, <16 x i16>) nounwind readnone
-
 define <8 x i32> @test_x86_avx2_phadd_d(<8 x i32> %a0, <8 x i32> %a1) {
 ; CHECK-LABEL: test_x86_avx2_phadd_d:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vphaddd %ymm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x02,0xc1]
 ; CHECK-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %res = call <8 x i32> @llvm.x86.avx2.phadd.d(<8 x i32> %a0, <8 x i32> %a1) ; <<8 x i32>> [#uses=1]
   ret <8 x i32> %res
 }

 declare <16 x i16> @llvm.x86.avx2.packusdw(<8 x i32>, <8 x i32>) nounwind readnone
 
 
 define <16 x i16> @test_x86_avx2_packusdw_fold() {
 ; X86-AVX-LABEL: test_x86_avx2_packusdw_fold:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [0,0,0,0,255,32767,65535,0,0,0,0,0,0,0,0,0]
 ; X86-AVX-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x05,A,A,A,A]
-; X86-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI51_0, kind: FK_Data_4
+; X86-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI47_0, kind: FK_Data_4
 ; X86-AVX-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X86-AVX512VL-LABEL: test_x86_avx2_packusdw_fold:
 ; X86-AVX512VL:       ## %bb.0:
-; X86-AVX512VL-NEXT:    vmovaps LCPI51_0, %ymm0 ## EVEX TO VEX Compression ymm0 = [0,0,0,0,255,32767,65535,0,0,0,0,0,0,0,0,0]
+; X86-AVX512VL-NEXT:    vmovaps LCPI47_0, %ymm0 ## EVEX TO VEX Compression ymm0 = [0,0,0,0,255,32767,65535,0,0,0,0,0,0,0,0,0]
 ; X86-AVX512VL-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x05,A,A,A,A]
-; X86-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI51_0, kind: FK_Data_4
+; X86-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI47_0, kind: FK_Data_4
 ; X86-AVX512VL-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X64-AVX-LABEL: test_x86_avx2_packusdw_fold:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [0,0,0,0,255,32767,65535,0,0,0,0,0,0,0,0,0]
 ; X64-AVX-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x05,A,A,A,A]
-; X64-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI51_0-4, kind: reloc_riprel_4byte
+; X64-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI47_0-4, kind: reloc_riprel_4byte
 ; X64-AVX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; X64-AVX512VL-LABEL: test_x86_avx2_packusdw_fold:
 ; X64-AVX512VL:       ## %bb.0:
 ; X64-AVX512VL-NEXT:    vmovaps {{.*}}(%rip), %ymm0 ## EVEX TO VEX Compression ymm0 = [0,0,0,0,255,32767,65535,0,0,0,0,0,0,0,0,0]
 ; X64-AVX512VL-NEXT:    ## encoding: [0xc5,0xfc,0x28,0x05,A,A,A,A]
-; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI51_0-4, kind: reloc_riprel_4byte
+; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI47_0-4, kind: reloc_riprel_4byte
 ; X64-AVX512VL-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i16> @llvm.x86.avx2.packusdw(<8 x i32> zeroinitializer, <8 x i32> <i32 255, i32 32767, i32 65535, i32 -1, i32 -32767, i32 -65535, i32 0, i32 -256>)
   ret <16 x i16> %res
 }
 
 
 define <32 x i8> @test_x86_avx2_pblendvb(<32 x i8> %a0, <32 x i8> %a1, <32 x i8> %a2) {
 ; CHECK-LABEL: test_x86_avx2_pblendvb:
; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   ret <4 x i32> %res
 }
 
 define <4 x i32> @test_x86_avx2_psrav_d_const(<4 x i32> %a0, <4 x i32> %a1) {
 ; X86-AVX-LABEL: test_x86_avx2_psrav_d_const:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vmovdqa {{.*#+}} xmm0 = [2,9,4294967284,23]
 ; X86-AVX-NEXT:    ## encoding: [0xc5,0xf9,0x6f,0x05,A,A,A,A]
-; X86-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI83_0, kind: FK_Data_4
-; X86-AVX-NEXT:    vpsravd LCPI83_1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x46,0x05,A,A,A,A]
-; X86-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI83_1, kind: FK_Data_4
+; X86-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI79_0, kind: FK_Data_4
+; X86-AVX-NEXT:    vpsravd LCPI79_1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x46,0x05,A,A,A,A]
+; X86-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI79_1, kind: FK_Data_4
 ; X86-AVX-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X86-AVX512VL-LABEL: test_x86_avx2_psrav_d_const:
 ; X86-AVX512VL:       ## %bb.0:
-; X86-AVX512VL-NEXT:    vmovdqa LCPI83_0, %xmm0 ## EVEX TO VEX Compression xmm0 = [2,9,4294967284,23]
+; X86-AVX512VL-NEXT:    vmovdqa LCPI79_0, %xmm0 ## EVEX TO VEX Compression xmm0 = [2,9,4294967284,23]
 ; X86-AVX512VL-NEXT:    ## encoding: [0xc5,0xf9,0x6f,0x05,A,A,A,A]
-; X86-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI83_0, kind: FK_Data_4
-; X86-AVX512VL-NEXT:    vpsravd LCPI83_1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x46,0x05,A,A,A,A]
-; X86-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI83_1, kind: FK_Data_4
+; X86-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI79_0, kind: FK_Data_4
+; X86-AVX512VL-NEXT:    vpsravd LCPI79_1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x46,0x05,A,A,A,A]
+; X86-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI79_1, kind: FK_Data_4
 ; X86-AVX512VL-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X64-AVX-LABEL: test_x86_avx2_psrav_d_const:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vmovdqa {{.*#+}} xmm0 = [2,9,4294967284,23]
 ; X64-AVX-NEXT:    ## encoding: [0xc5,0xf9,0x6f,0x05,A,A,A,A]
-; X64-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI83_0-4, kind: reloc_riprel_4byte
+; X64-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI79_0-4, kind: reloc_riprel_4byte
 ; X64-AVX-NEXT:    vpsravd {{.*}}(%rip), %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x46,0x05,A,A,A,A]
-; X64-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI83_1-4, kind: reloc_riprel_4byte
+; X64-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI79_1-4, kind: reloc_riprel_4byte
 ; X64-AVX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; X64-AVX512VL-LABEL: test_x86_avx2_psrav_d_const:
 ; X64-AVX512VL:       ## %bb.0:
 ; X64-AVX512VL-NEXT:    vmovdqa {{.*}}(%rip), %xmm0 ## EVEX TO VEX Compression xmm0 = [2,9,4294967284,23]
 ; X64-AVX512VL-NEXT:    ## encoding: [0xc5,0xf9,0x6f,0x05,A,A,A,A]
-; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI83_0-4, kind: reloc_riprel_4byte
+; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI79_0-4, kind: reloc_riprel_4byte
 ; X64-AVX512VL-NEXT:    vpsravd {{.*}}(%rip), %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x46,0x05,A,A,A,A]
-; X64-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI83_1-4, kind: reloc_riprel_4byte
+; X64-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI79_1-4, kind: reloc_riprel_4byte
 ; X64-AVX512VL-NEXT:    retq ## encoding: [0xc3]
   %res = call <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32> <i32 2, i32 9, i32 -12, i32 23>, <4 x i32> <i32 1, i32 18, i32 35, i32 52>)
   ret <4 x i32> %res
 }
 declare <4 x i32> @llvm.x86.avx2.psrav.d(<4 x i32>, <4 x i32>) nounwind readnone
 
 define <8 x i32> @test_x86_avx2_psrav_d_256(<8 x i32> %a0, <8 x i32> %a1) {
 ; AVX2-LABEL: test_x86_avx2_psrav_d_256:
; AVX512VL-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   ret <8 x i32> %res
 }
 
 define <8 x i32> @test_x86_avx2_psrav_d_256_const(<8 x i32> %a0, <8 x i32> %a1) {
 ; X86-AVX-LABEL: test_x86_avx2_psrav_d_256_const:
 ; X86-AVX:       ## %bb.0:
 ; X86-AVX-NEXT:    vmovdqa {{.*#+}} ymm0 = [2,9,4294967284,23,4294967270,37,4294967256,51]
 ; X86-AVX-NEXT:    ## encoding: [0xc5,0xfd,0x6f,0x05,A,A,A,A]
-; X86-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI85_0, kind: FK_Data_4
-; X86-AVX-NEXT:    vpsravd LCPI85_1, %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x46,0x05,A,A,A,A]
-; X86-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI85_1, kind: FK_Data_4
+; X86-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI81_0, kind: FK_Data_4
+; X86-AVX-NEXT:    vpsravd LCPI81_1, %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x46,0x05,A,A,A,A]
+; X86-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI81_1, kind: FK_Data_4
 ; X86-AVX-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X86-AVX512VL-LABEL: test_x86_avx2_psrav_d_256_const:
 ; X86-AVX512VL:       ## %bb.0:
-; X86-AVX512VL-NEXT:    vmovdqa LCPI85_0, %ymm0 ## EVEX TO VEX Compression ymm0 = [2,9,4294967284,23,4294967270,37,4294967256,51]
+; X86-AVX512VL-NEXT:    vmovdqa LCPI81_0, %ymm0 ## EVEX TO VEX Compression ymm0 = [2,9,4294967284,23,4294967270,37,4294967256,51]
 ; X86-AVX512VL-NEXT:    ## encoding: [0xc5,0xfd,0x6f,0x05,A,A,A,A]
-; X86-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI85_0, kind: FK_Data_4
-; X86-AVX512VL-NEXT:    vpsravd LCPI85_1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7d,0x46,0x05,A,A,A,A]
-; X86-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI85_1, kind: FK_Data_4
+; X86-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI81_0, kind: FK_Data_4
+; X86-AVX512VL-NEXT:    vpsravd LCPI81_1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7d,0x46,0x05,A,A,A,A]
+; X86-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI81_1, kind: FK_Data_4
 ; X86-AVX512VL-NEXT:    retl ## encoding: [0xc3]
 ;
 ; X64-AVX-LABEL: test_x86_avx2_psrav_d_256_const:
 ; X64-AVX:       ## %bb.0:
 ; X64-AVX-NEXT:    vmovdqa {{.*#+}} ymm0 = [2,9,4294967284,23,4294967270,37,4294967256,51]
 ; X64-AVX-NEXT:    ## encoding: [0xc5,0xfd,0x6f,0x05,A,A,A,A]
-; X64-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI85_0-4, kind: reloc_riprel_4byte
+; X64-AVX-NEXT:    ## fixup A - offset: 4, value: LCPI81_0-4, kind: reloc_riprel_4byte
 ; X64-AVX-NEXT:    vpsravd {{.*}}(%rip), %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x46,0x05,A,A,A,A]
-; X64-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI85_1-4, kind: reloc_riprel_4byte
+; X64-AVX-NEXT:    ## fixup A - offset: 5, value: LCPI81_1-4, kind: reloc_riprel_4byte
 ; X64-AVX-NEXT:    retq ## encoding: [0xc3]
 ;
 ; X64-AVX512VL-LABEL: test_x86_avx2_psrav_d_256_const:
 ; X64-AVX512VL:       ## %bb.0:
 ; X64-AVX512VL-NEXT:    vmovdqa {{.*}}(%rip), %ymm0 ## EVEX TO VEX Compression ymm0 = [2,9,4294967284,23,4294967270,37,4294967256,51]
 ; X64-AVX512VL-NEXT:    ## encoding: [0xc5,0xfd,0x6f,0x05,A,A,A,A]
-; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI85_0-4, kind: reloc_riprel_4byte
+; X64-AVX512VL-NEXT:    ## fixup A - offset: 4, value: LCPI81_0-4, kind: reloc_riprel_4byte
 ; X64-AVX512VL-NEXT:    vpsravd {{.*}}(%rip), %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x7d,0x46,0x05,A,A,A,A]
-; X64-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI85_1-4, kind: reloc_riprel_4byte
+; X64-AVX512VL-NEXT:    ## fixup A - offset: 5, value: LCPI81_1-4, kind: reloc_riprel_4byte
 ; X64-AVX512VL-NEXT:    retq ## encoding: [0xc3]
   %res = call <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32> <i32 2, i32 9, i32 -12, i32 23, i32 -26, i32 37, i32 -40, i32 51>, <8 x i32> <i32 1, i32 18, i32 35, i32 52, i32 69, i32 15, i32 32, i32 49>)
   ret <8 x i32> %res
 }
 declare <8 x i32> @llvm.x86.avx2.psrav.d.256(<8 x i32>, <8 x i32>) nounwind readnone
 
 define <2 x double> @test_x86_avx2_gather_d_pd(<2 x double> %a0, i8* %a1, <4 x i32> %idx, <2 x double> %mask) {
 ; X86-LABEL: test_x86_avx2_gather_d_pd:

test/CodeGen/X86/avx512bw-intrinsics-canonical.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512BW
 ; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512F-32
 
-; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/sse2-builtins.c
+; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512bw-builtins.c
 
+;
+; Signed Saturation
+;
+
+define <32 x i16> @test_mask_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rr_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rr_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %res = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %res
+}
+declare <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_mask_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rrk_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %edi, %k1
+; AVX512BW-NEXT:    vpaddsw %zmm1, %zmm0, %zmm2 {%k1}
+; AVX512BW-NEXT:    vmovdqa64 %zmm2, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rrk_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpaddsw %zmm1, %zmm0, %zmm2 {%k1}
+; AVX512F-32-NEXT:    vmovdqa64 %zmm2, %zmm0
+; AVX512F-32-NEXT:    retl
+  %1 = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rrkz_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %edi, %k1
+; AVX512BW-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rrkz_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-32-NEXT:    retl
+  %1 = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_adds_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rm_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    vpaddsw (%rdi), %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rm_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    vpaddsw (%eax), %zmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_mask_adds_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rmk_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %esi, %k1
+; AVX512BW-NEXT:    vpaddsw (%rdi), %zmm0, %zmm1 {%k1}
+; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rmk_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpaddsw (%eax), %zmm0, %zmm1 {%k1}
+; AVX512F-32-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512F-32-NEXT:    retl
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_adds_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rmkz_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %esi, %k1
+; AVX512BW-NEXT:    vpaddsw (%rdi), %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rmkz_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpaddsw (%eax), %zmm0, %zmm0 {%k1} {z}
+; AVX512F-32-NEXT:    retl
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_mask_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rr_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rr_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %sub = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %sub
+}
+declare <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16>, <32 x i16>)
+
+define <32 x i16> @test_mask_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rrk_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %edi, %k1
+; AVX512BW-NEXT:    vpsubsw %zmm1, %zmm0, %zmm2 {%k1}
+; AVX512BW-NEXT:    vmovdqa64 %zmm2, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rrk_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpsubsw %zmm1, %zmm0, %zmm2 {%k1}
+; AVX512F-32-NEXT:    vmovdqa64 %zmm2, %zmm0
+; AVX512F-32-NEXT:    retl
+  %sub = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i16> %sub, <32 x i16> %passThru
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rrkz_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %edi, %k1
+; AVX512BW-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rrkz_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-32-NEXT:    retl
+  %sub = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i16> %sub, <32 x i16> zeroinitializer
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rm_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    vpsubsw (%rdi), %zmm0, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rm_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    vpsubsw (%eax), %zmm0, %zmm0
+; AVX512F-32-NEXT:    retl
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %sub = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %sub
+}
+
+define <32 x i16> @test_mask_subs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rmk_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %esi, %k1
+; AVX512BW-NEXT:    vpsubsw (%rdi), %zmm0, %zmm1 {%k1}
+; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rmk_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpsubsw (%eax), %zmm0, %zmm1 {%k1}
+; AVX512F-32-NEXT:    vmovdqa64 %zmm1, %zmm0
+; AVX512F-32-NEXT:    retl
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %sub = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i16> %sub, <32 x i16> %passThru
+  ret <32 x i16> %res
+}
+
+define <32 x i16> @test_mask_subs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rmkz_512:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    kmovd %esi, %k1
+; AVX512BW-NEXT:    vpsubsw (%rdi), %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rmkz_512:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; AVX512F-32-NEXT:    kmovd {{[0-9]+}}(%esp), %k1
+; AVX512F-32-NEXT:    vpsubsw (%eax), %zmm0, %zmm0 {%k1} {z}
+; AVX512F-32-NEXT:    retl
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %sub = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %a, <32 x i16> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i16> %sub, <32 x i16> zeroinitializer
+  ret <32 x i16> %res
+}
+
+
+define <64 x i16> @test_mask_adds_epi16_rr_1024(<64 x i16> %a, <64 x i16> %b) {
+; AVX512BW-LABEL: test_mask_adds_epi16_rr_1024:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    vpaddsw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpaddsw %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_adds_epi16_rr_1024:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    pushl %ebp
+; AVX512F-32-NEXT:    .cfi_def_cfa_offset 8
+; AVX512F-32-NEXT:    .cfi_offset %ebp, -8
+; AVX512F-32-NEXT:    movl %esp, %ebp
+; AVX512F-32-NEXT:    .cfi_def_cfa_register %ebp
+; AVX512F-32-NEXT:    andl $-64, %esp
+; AVX512F-32-NEXT:    subl $64, %esp
+; AVX512F-32-NEXT:    vpaddsw %zmm2, %zmm0, %zmm0
+; AVX512F-32-NEXT:    vpaddsw 8(%ebp), %zmm1, %zmm1
+; AVX512F-32-NEXT:    movl %ebp, %esp
+; AVX512F-32-NEXT:    popl %ebp
+; AVX512F-32-NEXT:    .cfi_def_cfa %esp, 4
+; AVX512F-32-NEXT:    retl
+  %1 = call <64 x i16> @llvm.sadd.sat.v64i16(<64 x i16> %a, <64 x i16> %b)
+  ret <64 x i16> %1
+}
+declare <64 x i16> @llvm.sadd.sat.v64i16(<64 x i16>, <64 x i16>)
+
+define <64 x i16> @test_mask_subs_epi16_rr_1024(<64 x i16> %a, <64 x i16> %b) {
+; AVX512BW-LABEL: test_mask_subs_epi16_rr_1024:
+; AVX512BW:       ## %bb.0:
+; AVX512BW-NEXT:    vpsubsw %zmm2, %zmm0, %zmm0
+; AVX512BW-NEXT:    vpsubsw %zmm3, %zmm1, %zmm1
+; AVX512BW-NEXT:    retq
+;
+; AVX512F-32-LABEL: test_mask_subs_epi16_rr_1024:
+; AVX512F-32:       # %bb.0:
+; AVX512F-32-NEXT:    pushl %ebp
+; AVX512F-32-NEXT:    .cfi_def_cfa_offset 8
+; AVX512F-32-NEXT:    .cfi_offset %ebp, -8
+; AVX512F-32-NEXT:    movl %esp, %ebp
+; AVX512F-32-NEXT:    .cfi_def_cfa_register %ebp
+; AVX512F-32-NEXT:    andl $-64, %esp
+; AVX512F-32-NEXT:    subl $64, %esp
+; AVX512F-32-NEXT:    vpsubsw %zmm2, %zmm0, %zmm0
+; AVX512F-32-NEXT:    vpsubsw 8(%ebp), %zmm1, %zmm1
+; AVX512F-32-NEXT:    movl %ebp, %esp
+; AVX512F-32-NEXT:    popl %ebp
+; AVX512F-32-NEXT:    .cfi_def_cfa %esp, 4
+; AVX512F-32-NEXT:    retl
+  %sub = call <64 x i16> @llvm.ssub.sat.v64i16(<64 x i16> %a, <64 x i16> %b)
+  ret <64 x i16> %sub
+}
+declare <64 x i16> @llvm.ssub.sat.v64i16(<64 x i16>, <64 x i16>);
+
+;
+; Unsigned Saturation
+;
 
 define <32 x i16> @test_mask_adds_epu16_rr_512(<32 x i16> %a, <32 x i16> %b) {
 ; AVX512BW-LABEL: test_mask_adds_epu16_rr_512:
 ; AVX512BW:       ## %bb.0:
 ; AVX512BW-NEXT:    vpaddusw %zmm1, %zmm0, %zmm0
 ; AVX512BW-NEXT:    retq
 ;
 ; AVX512F-32-LABEL: test_mask_adds_epu16_rr_512:

test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll

 ; X64-NEXT:    retq # encoding: [0xc3]
   %b = load <64 x i8>, <64 x i8>* %ptr_b
   %res = call <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8> %a, <64 x i8> %b, <64 x i8> zeroinitializer, i64 %mask)
   ret <64 x i8> %res
 }
 
 declare <64 x i8> @llvm.x86.avx512.mask.psubus.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64)
 
+define <32 x i16> @test_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
+; CHECK-LABEL: test_adds_epi16_rr_512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_adds_epi16_rrk_512:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
+; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rrk_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
+; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
+; X86-LABEL: test_adds_epi16_rrkz_512:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rrkz_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_adds_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
+; X86-LABEL: test_adds_epi16_rm_512:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vpaddsw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rm_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpaddsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_adds_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_adds_epi16_rmk_512:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0x08]
+; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rmk_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0x0f]
+; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_adds_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
+; X86-LABEL: test_adds_epi16_rmkz_512:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rmkz_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16>, <32 x i16>)
+
 define <32 x i16> @test_mask_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
 ; CHECK-LABEL: test_mask_adds_epi16_rr_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0xc1]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %res = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %res
 }

 ; X64-NEXT:    retq # encoding: [0xc3]
   %b = load <32 x i16>, <32 x i16>* %ptr_b
   %res = call <32 x i16> @llvm.x86.avx512.mask.padds.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.padds.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32)
 
+define <32 x i16> @test_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
+; CHECK-LABEL: test_subs_epi16_rr_512:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0xc1]
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_subs_epi16_rrk_512:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
+; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_subs_epi16_rrk_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
+; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
+; X86-LABEL: test_subs_epi16_rrkz_512:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_subs_epi16_rrkz_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_subs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
+; X86-LABEL: test_subs_epi16_rm_512:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vpsubsw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_subs_epi16_rm_512:
+; X64:       # %bb.0:
+; X64-NEXT:    vpsubsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @test_subs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
+; X86-LABEL: test_subs_epi16_rmk_512:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0x08]
+; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_subs_epi16_rmk_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0x0f]
+; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
+  ret <32 x i16> %3
+}
+
+define <32 x i16> @test_subs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
+; X86-LABEL: test_subs_epi16_rmkz_512:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_subs_epi16_rmkz_512:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i16>, <32 x i16>* %ptr_b
+  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
+  ret <32 x i16> %3
+}
+
+declare <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16>, <32 x i16>)
+
 define <32 x i16> @test_mask_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
 ; CHECK-LABEL: test_mask_subs_epi16_rr_512:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0xc1]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %res = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> %a, <32 x i16> %b, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %res
 }

test/CodeGen/X86/avx512bw-intrinsics.ll

; X64-NEXT:    retq # encoding: [0xc3]
   %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
   %2 = bitcast i64 %mask to <64 x i1>
   %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
   ret <64 x i8> %3
 }
 
 declare <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16>, <32 x i16>)
 
-define <32 x i16> @test_mask_adds_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: test_mask_adds_epi16_rr_512:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0xc1]
-; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
-  ret <32 x i16> %1
-}
-
-define <32 x i16> @test_mask_adds_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rrk_512:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
-; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rrk_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0xd1]
-; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
-  ret <32 x i16> %3
-}
-
-define <32 x i16> @test_mask_adds_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rrkz_512:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rrkz_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
-  ret <32 x i16> %3
-}
-
-define <32 x i16> @test_mask_adds_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
-; X86-LABEL: test_mask_adds_epi16_rm_512:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vpaddsw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rm_512:
-; X64:       # %bb.0:
-; X64-NEXT:    vpaddsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xed,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i16>, <32 x i16>* %ptr_b
-  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
-  ret <32 x i16> %1
-}
-
-define <32 x i16> @test_mask_adds_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rmk_512:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0x08]
-; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rmk_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xed,0x0f]
-; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i16>, <32 x i16>* %ptr_b
-  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
-  ret <32 x i16> %3
-}
-
-define <32 x i16> @test_mask_adds_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rmkz_512:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rmkz_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xed,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i16>, <32 x i16>* %ptr_b
-  %1 = call <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
-  ret <32 x i16> %3
-}
-
-declare <32 x i16> @llvm.x86.avx512.padds.w.512(<32 x i16>, <32 x i16>)
-
-define <32 x i16> @test_mask_subs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: test_mask_subs_epi16_rr_512:
-; CHECK:       # %bb.0:
-; CHECK-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0xc1]
-; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
-  ret <32 x i16> %1
-}
-
-define <32 x i16> @test_mask_subs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <32 x i16> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rrk_512:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
-; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rrk_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0xd1]
-; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
-  ret <32 x i16> %3
-}
-
-define <32 x i16> @test_mask_subs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rrkz_512:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rrkz_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
-  ret <32 x i16> %3
-}
-
-define <32 x i16> @test_mask_subs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
-; X86-LABEL: test_mask_subs_epi16_rm_512:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vpsubsw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rm_512:
-; X64:       # %bb.0:
-; X64-NEXT:    vpsubsw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe9,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i16>, <32 x i16>* %ptr_b
-  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
-  ret <32 x i16> %1
-}
-
-define <32 x i16> @test_mask_subs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rmk_512:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0x08]
-; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rmk_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe9,0x0f]
-; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i16>, <32 x i16>* %ptr_b
-  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
-  ret <32 x i16> %3
-}
-
-define <32 x i16> @test_mask_subs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rmkz_512:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rmkz_512:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe9,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i16>, <32 x i16>* %ptr_b
-  %1 = call <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16> %a, <32 x i16> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
-  ret <32 x i16> %3
-}
-
-declare <32 x i16> @llvm.x86.avx512.psubs.w.512(<32 x i16>, <32 x i16>)
-
-
 define <32 x i16>@test_int_x86_avx512_mask_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpermt2var_hi_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
 ; X86-NEXT:    vpermt2w %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x7d,0xda]
 ; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
 ; X86-NEXT:    vpermt2w %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x7d,0xca]
 ; X86-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]

test/CodeGen/X86/avx512bwvl-intrinsics-canonical.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl -mattr=+avx512bw -mattr=+avx512vl --show-mc-encoding| FileCheck %s
 
-; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/sse2-builtins.c
+; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512vlbw-builtins.c
 
 
+;
+; Signed Saturation
+;
+
+define <8 x i16> @test_mask_adds_epi16_rr_128(<8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: test_mask_adds_epi16_rr_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xed,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %1
+}
+declare <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16>, <8 x i16>)
+
+define <8 x i16> @test_mask_adds_epi16_rrk_128(<8 x i16> %a, <8 x i16> %b, <8 x i16> %passThru, i8 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rrk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsw %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xed,0xd1]
+; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_mask_adds_epi16_rrkz_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rrkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xed,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_mask_adds_epi16_rm_128(<8 x i16> %a, <8 x i16>* %ptr_b) {
+; CHECK-LABEL: test_mask_adds_epi16_rm_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsw (%rdi), %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xed,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %1
+}
+
+define <8 x i16> @test_mask_adds_epi16_rmk_128(<8 x i16> %a, <8 x i16>* %ptr_b, <8 x i16> %passThru, i8 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rmk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsw (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xed,0x0f]
+; CHECK-NEXT:    vmovdqa %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_mask_adds_epi16_rmkz_128(<8 x i16> %a, <8 x i16>* %ptr_b, i8 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rmkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsw (%rdi), %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xed,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
+  ret <8 x i16> %3
+}
+
+define <16 x i16> @test_mask_adds_epi16_rr_256(<16 x i16> %a, <16 x i16> %b) {
+; CHECK-LABEL: test_mask_adds_epi16_rr_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xed,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  ret <16 x i16> %1
+}
+declare <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16>, <16 x i16>)
+
+define <16 x i16> @test_mask_adds_epi16_rrk_256(<16 x i16> %a, <16 x i16> %b, <16 x i16> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rrk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsw %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xed,0xd1]
+; CHECK-NEXT:    vmovdqa %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_mask_adds_epi16_rrkz_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rrkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xed,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_mask_adds_epi16_rm_256(<16 x i16> %a, <16 x i16>* %ptr_b) {
+; CHECK-LABEL: test_mask_adds_epi16_rm_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsw (%rdi), %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xed,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  ret <16 x i16> %1
+}
+
+define <16 x i16> @test_mask_adds_epi16_rmk_256(<16 x i16> %a, <16 x i16>* %ptr_b, <16 x i16> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rmk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsw (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xed,0x0f]
+; CHECK-NEXT:    vmovdqa %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_mask_adds_epi16_rmkz_256(<16 x i16> %a, <16 x i16>* %ptr_b, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi16_rmkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsw (%rdi), %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xed,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
+  ret <16 x i16> %3
+}
+
+define <8 x i16> @test_mask_subs_epi16_rr_128(<8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: test_mask_subs_epi16_rr_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe9,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %sub
+}
+declare <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16>, <8 x i16>)
+
+define <8 x i16> @test_mask_subs_epi16_rrk_128(<8 x i16> %a, <8 x i16> %b, <8 x i16> %passThru, i8 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rrk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsw %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xe9,0xd1]
+; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %bc = bitcast i8 %mask to <8 x i1>
+  %res = select <8 x i1> %bc, <8 x i16> %sub, <8 x i16> %passThru
+  ret <8 x i16> %res
+}
+
+define <8 x i16> @test_mask_subs_epi16_rrkz_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rrkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xe9,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %bc = bitcast i8 %mask to <8 x i1>
+  %res = select <8 x i1> %bc, <8 x i16> %sub, <8 x i16> zeroinitializer
+  ret <8 x i16> %res
+}
+
+define <8 x i16> @test_mask_subs_epi16_rm_128(<8 x i16> %a, <8 x i16>* %ptr_b) {
+; CHECK-LABEL: test_mask_subs_epi16_rm_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsw (%rdi), %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe9,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %sub = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %sub
+}
+
+define <8 x i16> @test_mask_subs_epi16_rmk_128(<8 x i16> %a, <8 x i16>* %ptr_b, <8 x i16> %passThru, i8 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rmk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsw (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xe9,0x0f]
+; CHECK-NEXT:    vmovdqa %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %sub = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %bc = bitcast i8 %mask to <8 x i1>
+  %res = select <8 x i1> %bc, <8 x i16> %sub, <8 x i16> %passThru
+  ret <8 x i16> %res
+}
+
+define <8 x i16> @test_mask_subs_epi16_rmkz_128(<8 x i16> %a, <8 x i16>* %ptr_b, i8 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rmkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsw (%rdi), %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xe9,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %sub = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  %bc = bitcast i8 %mask to <8 x i1>
+  %res = select <8 x i1> %bc, <8 x i16> %sub, <8 x i16> zeroinitializer
+  ret <8 x i16> %res
+}
+
+define <16 x i16> @test_mask_subs_epi16_rr_256(<16 x i16> %a, <16 x i16> %b) {
+; CHECK-LABEL: test_mask_subs_epi16_rr_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe9,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  ret <16 x i16> %sub
+}
+declare <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16>, <16 x i16>)
+
+define <16 x i16> @test_mask_subs_epi16_rrk_256(<16 x i16> %a, <16 x i16> %b, <16 x i16> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rrk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsw %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xe9,0xd1]
+; CHECK-NEXT:    vmovdqa %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i16> %sub, <16 x i16> %passThru
+  ret <16 x i16> %res
+}
+
+define <16 x i16> @test_mask_subs_epi16_rrkz_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rrkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i16> %sub, <16 x i16> zeroinitializer
+  ret <16 x i16> %res
+}
+
+define <16 x i16> @test_mask_subs_epi16_rm_256(<16 x i16> %a, <16 x i16>* %ptr_b) {
+; CHECK-LABEL: test_mask_subs_epi16_rm_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsw (%rdi), %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe9,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %sub = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  ret <16 x i16> %sub
+}
+
+define <16 x i16> @test_mask_subs_epi16_rmk_256(<16 x i16> %a, <16 x i16>* %ptr_b, <16 x i16> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rmk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsw (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xe9,0x0f]
+; CHECK-NEXT:    vmovdqa %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %sub = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i16> %sub, <16 x i16> %passThru
+  ret <16 x i16> %res
+}
+
+define <16 x i16> @test_mask_subs_epi16_rmkz_256(<16 x i16> %a, <16 x i16>* %ptr_b, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi16_rmkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsw (%rdi), %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %sub = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %a, <16 x i16> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i16> %sub, <16 x i16> zeroinitializer
+  ret <16 x i16> %res
+}
+
+define <16 x i8> @test_mask_adds_epi8_rr_128(<16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: test_mask_adds_epi8_rr_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %1
+}
+declare <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8>, <16 x i8>)
+
+define <16 x i8> @test_mask_adds_epi8_rrk_128(<16 x i8> %a, <16 x i8> %b, <16 x i8> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rrk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsb %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xec,0xd1]
+; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_mask_adds_epi8_rrkz_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rrkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xec,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_mask_adds_epi8_rm_128(<16 x i8> %a, <16 x i8>* %ptr_b) {
+; CHECK-LABEL: test_mask_adds_epi8_rm_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %1
+}
+
+define <16 x i8> @test_mask_adds_epi8_rmk_128(<16 x i8> %a, <16 x i8>* %ptr_b, <16 x i8> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rmk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsb (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xec,0x0f]
+; CHECK-NEXT:    vmovdqa %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_mask_adds_epi8_rmkz_128(<16 x i8> %a, <16 x i8>* %ptr_b, i16 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rmkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xec,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
+  ret <16 x i8> %3
+}
+
+define <32 x i8> @test_mask_adds_epi8_rr_256(<32 x i8> %a, <32 x i8> %b) {
+; CHECK-LABEL: test_mask_adds_epi8_rr_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xec,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  ret <32 x i8> %1
+}
+declare <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8>, <32 x i8>)
+
+define <32 x i8> @test_mask_adds_epi8_rrk_256(<32 x i8> %a, <32 x i8> %b, <32 x i8> %passThru, i32 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rrk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsb %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xec,0xd1]
+; CHECK-NEXT:    vmovdqa %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_mask_adds_epi8_rrkz_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rrkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xec,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_mask_adds_epi8_rm_256(<32 x i8> %a, <32 x i8>* %ptr_b) {
+; CHECK-LABEL: test_mask_adds_epi8_rm_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpaddsb (%rdi), %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xec,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  ret <32 x i8> %1
+}
+
+define <32 x i8> @test_mask_adds_epi8_rmk_256(<32 x i8> %a, <32 x i8>* %ptr_b, <32 x i8> %passThru, i32 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rmk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsb (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xec,0x0f]
+; CHECK-NEXT:    vmovdqa %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_mask_adds_epi8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
+; CHECK-LABEL: test_mask_adds_epi8_rmkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpaddsb (%rdi), %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xec,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
+define <16 x i8> @test_mask_subs_epi8_rr_128(<16 x i8> %a, <16 x i8> %b) {
+; CHECK-LABEL: test_mask_subs_epi8_rr_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe8,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %sub
+}
+declare <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8>, <16 x i8>)
+
+define <16 x i8> @test_mask_subs_epi8_rrk_128(<16 x i8> %a, <16 x i8> %b, <16 x i8> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rrk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsb %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xe8,0xd1]
+; CHECK-NEXT:    vmovdqa %xmm2, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i8> %sub, <16 x i8> %passThru
+  ret <16 x i8> %res
+}
+
+define <16 x i8> @test_mask_subs_epi8_rrkz_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rrkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xe8,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i8> %sub, <16 x i8> zeroinitializer
+  ret <16 x i8> %res
+}
+
+define <16 x i8> @test_mask_subs_epi8_rm_128(<16 x i8> %a, <16 x i8>* %ptr_b) {
+; CHECK-LABEL: test_mask_subs_epi8_rm_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsb (%rdi), %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe8,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %sub = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %sub
+}
+
+define <16 x i8> @test_mask_subs_epi8_rmk_128(<16 x i8> %a, <16 x i8>* %ptr_b, <16 x i8> %passThru, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rmk_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsb (%rdi), %xmm0, %xmm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x09,0xe8,0x0f]
+; CHECK-NEXT:    vmovdqa %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %sub = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i8> %sub, <16 x i8> %passThru
+  ret <16 x i8> %res
+}
+
+define <16 x i8> @test_mask_subs_epi8_rmkz_128(<16 x i8> %a, <16 x i8>* %ptr_b, i16 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rmkz_128:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsb (%rdi), %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0x89,0xe8,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %sub = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  %bc = bitcast i16 %mask to <16 x i1>
+  %res = select <16 x i1> %bc, <16 x i8> %sub, <16 x i8> zeroinitializer
+  ret <16 x i8> %res
+}
+
+define <32 x i8> @test_mask_subs_epi8_rr_256(<32 x i8> %a, <32 x i8> %b) {
+; CHECK-LABEL: test_mask_subs_epi8_rr_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe8,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  ret <32 x i8> %sub
+}
+declare <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8>, <32 x i8>)
+
+define <32 x i8> @test_mask_subs_epi8_rrk_256(<32 x i8> %a, <32 x i8> %b, <32 x i8> %passThru, i32 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rrk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsb %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xe8,0xd1]
+; CHECK-NEXT:    vmovdqa %ymm2, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i8> %sub, <32 x i8> %passThru
+  ret <32 x i8> %res
+}
+
+define <32 x i8> @test_mask_subs_epi8_rrkz_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rrkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
+; CHECK-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %sub = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i8> %sub, <32 x i8> zeroinitializer
+  ret <32 x i8> %res
+}
+
+define <32 x i8> @test_mask_subs_epi8_rm_256(<32 x i8> %a, <32 x i8>* %ptr_b) {
+; CHECK-LABEL: test_mask_subs_epi8_rm_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    vpsubsb (%rdi), %ymm0, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0xe8,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %sub = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  ret <32 x i8> %sub
+}
+
+define <32 x i8> @test_mask_subs_epi8_rmk_256(<32 x i8> %a, <32 x i8>* %ptr_b, <32 x i8> %passThru, i32 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rmk_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsb (%rdi), %ymm0, %ymm1 {%k1} ## encoding: [0x62,0xf1,0x7d,0x29,0xe8,0x0f]
+; CHECK-NEXT:    vmovdqa %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %sub = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i8> %sub, <32 x i8> %passThru
+  ret <32 x i8> %res
+}
+
+define <32 x i8> @test_mask_subs_epi8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
+; CHECK-LABEL: test_mask_subs_epi8_rmkz_256:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    kmovd %esi, %k1 ## encoding: [0xc5,0xfb,0x92,0xce]
+; CHECK-NEXT:    vpsubsb (%rdi), %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0x07]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %sub = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %a, <32 x i8> %b)
+  %bc = bitcast i32 %mask to <32 x i1>
+  %res = select <32 x i1> %bc, <32 x i8> %sub, <32 x i8> zeroinitializer
+  ret <32 x i8> %res
+}
+
+;
+; Unsigned Saturation
+;
+
 define <8 x i16> @test_mask_adds_epu16_rr_128(<8 x i16> %a, <8 x i16> %b) {
 ; CHECK-LABEL: test_mask_adds_epu16_rr_128:
 ; CHECK:       ## %bb.0:
 ; CHECK-NEXT:    vpaddusw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xdd,0xc1]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %1 = call <8 x i16> @llvm.uadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
   ret <8 x i16> %1
 }

test/CodeGen/X86/avx512bwvl-intrinsics-upgrade.ll

 ; X64-NEXT:    retq # encoding: [0xc3]
   %b = load <16 x i16>, <16 x i16>* %ptr_b
   %res = call <16 x i16> @llvm.x86.avx512.mask.padds.w.256(<16 x i16> %a, <16 x i16> %b, <16 x i16> zeroinitializer, i16 %mask)
   ret <16 x i16> %res
 }
 
 declare <16 x i16> @llvm.x86.avx512.mask.padds.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
 
+declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) #0
+
+define <8 x i16> @test_test_subs_epi16_rrk_128(<8 x i16> %a, <8 x i16> %b, <8 x i16> %passThru, i8 %mask) {
+; X86-LABEL: test_test_subs_epi16_rrk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
+; X86-NEXT:    vpsubsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0xd1]
+; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rrk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0xd1]
+; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_test_subs_epi16_rrkz_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
+; X86-LABEL: test_test_subs_epi16_rrkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
+; X86-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rrkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_test_subs_epi16_rmk_128(<8 x i16> %a, <8 x i16>* %ptr_b, <8 x i16> %passThru, i8 %mask) {
+; X86-LABEL: test_test_subs_epi16_rmk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
+; X86-NEXT:    vpsubsw (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0x08]
+; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rmk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsw (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0x0f]
+; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_test_subs_epi16_rmkz_128(<8 x i16> %a, <8 x i16>* %ptr_b, i8 %mask) {
+; X86-LABEL: test_test_subs_epi16_rmkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
+; X86-NEXT:    vpsubsw (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rmkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsw (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
+  ret <8 x i16> %3
+}
+
+declare <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16>, <16 x i16>) #0
+
+define <16 x i16> @test_test_subs_epi16_rrk_256(<16 x i16> %a, <16 x i16> %b, <16 x i16> %passThru, i16 %mask) {
+; X86-LABEL: test_test_subs_epi16_rrk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0xd1]
+; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rrk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0xd1]
+; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_test_subs_epi16_rrkz_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
+; X86-LABEL: test_test_subs_epi16_rrkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rrkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_test_subs_epi16_rmk_256(<16 x i16> %a, <16 x i16>* %ptr_b, <16 x i16> %passThru, i16 %mask) {
+; X86-LABEL: test_test_subs_epi16_rmk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsw (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0x08]
+; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rmk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsw (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0x0f]
+; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_test_subs_epi16_rmkz_256(<16 x i16> %a, <16 x i16>* %ptr_b, i16 %mask) {
+; X86-LABEL: test_test_subs_epi16_rmkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsw (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi16_rmkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsw (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
+  ret <16 x i16> %3
+}
+
+declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) #0
+
+define <16 x i8> @test_test_subs_epi8_rrk_128(<16 x i8> %a, <16 x i8> %b, <16 x i8> %passThru, i16 %mask) {
+; X86-LABEL: test_test_subs_epi8_rrk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0xd1]
+; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rrk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0xd1]
+; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_test_subs_epi8_rrkz_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
+; X86-LABEL: test_test_subs_epi8_rrkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rrkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_test_subs_epi8_rmk_128(<16 x i8> %a, <16 x i8>* %ptr_b, <16 x i8> %passThru, i16 %mask) {
+; X86-LABEL: test_test_subs_epi8_rmk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsb (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0x08]
+; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rmk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsb (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0x0f]
+; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_test_subs_epi8_rmkz_128(<16 x i8> %a, <16 x i8>* %ptr_b, i16 %mask) {
+; X86-LABEL: test_test_subs_epi8_rmkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsb (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rmkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsb (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
+  ret <16 x i8> %3
+}
+
+declare <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8>, <32 x i8>) #0
+
+define <32 x i8> @test_test_subs_epi8_rrk_256(<32 x i8> %a, <32 x i8> %b, <32 x i8> %passThru, i32 %mask) {
+; X86-LABEL: test_test_subs_epi8_rrk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0xd1]
+; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rrk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0xd1]
+; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_test_subs_epi8_rrkz_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
+; X86-LABEL: test_test_subs_epi8_rrkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rrkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_test_subs_epi8_rmk_256(<32 x i8> %a, <32 x i8>* %ptr_b, <32 x i8> %passThru, i32 %mask) {
+; X86-LABEL: test_test_subs_epi8_rmk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsb (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0x08]
+; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rmk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsb (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0x0f]
+; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_test_subs_epi8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
+; X86-LABEL: test_test_subs_epi8_rmkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpsubsb (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_test_subs_epi8_rmkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpsubsb (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
 define <8 x i16> @test_mask_subs_epi16_rr_128(<8 x i16> %a, <8 x i16> %b) {
 ; CHECK-LABEL: test_mask_subs_epi16_rr_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe9,0xc1]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %res = call <8 x i16> @llvm.x86.avx512.mask.psubs.w.128(<8 x i16> %a, <8 x i16> %b, <8 x i16> zeroinitializer, i8 -1)
   ret <8 x i16> %res
 }

 ; X64-NEXT:    retq # encoding: [0xc3]
   %b = load <16 x i16>, <16 x i16>* %ptr_b
   %res = call <16 x i16> @llvm.x86.avx512.mask.psubs.w.256(<16 x i16> %a, <16 x i16> %b, <16 x i16> zeroinitializer, i16 %mask)
   ret <16 x i16> %res
 }
 
 declare <16 x i16> @llvm.x86.avx512.mask.psubs.w.256(<16 x i16>, <16 x i16>, <16 x i16>, i16)
 
+declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) #0
+
+define <8 x i16> @test_adds_epi16_rrk_128(<8 x i16> %a, <8 x i16> %b, <8 x i16> %passThru, i8 %mask) {
+; X86-LABEL: test_adds_epi16_rrk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
+; X86-NEXT:    vpaddsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0xd1]
+; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rrk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0xd1]
+; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_adds_epi16_rrkz_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
+; X86-LABEL: test_adds_epi16_rrkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
+; X86-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rrkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_adds_epi16_rmk_128(<8 x i16> %a, <8 x i16>* %ptr_b, <8 x i16> %passThru, i8 %mask) {
+; X86-LABEL: test_adds_epi16_rmk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
+; X86-NEXT:    vpaddsw (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0x08]
+; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rmk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsw (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0x0f]
+; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
+  ret <8 x i16> %3
+}
+
+define <8 x i16> @test_adds_epi16_rmkz_128(<8 x i16> %a, <8 x i16>* %ptr_b, i8 %mask) {
+; X86-LABEL: test_adds_epi16_rmkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
+; X86-NEXT:    vpaddsw (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rmkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsw (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <8 x i16>, <8 x i16>* %ptr_b
+  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
+  ret <8 x i16> %3
+}
+
+declare <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16>, <16 x i16>) #0
+
+define <16 x i16> @test_adds_epi16_rrk_256(<16 x i16> %a, <16 x i16> %b, <16 x i16> %passThru, i16 %mask) {
+; X86-LABEL: test_adds_epi16_rrk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0xd1]
+; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rrk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0xd1]
+; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_adds_epi16_rrkz_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
+; X86-LABEL: test_adds_epi16_rrkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rrkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_adds_epi16_rmk_256(<16 x i16> %a, <16 x i16>* %ptr_b, <16 x i16> %passThru, i16 %mask) {
+; X86-LABEL: test_adds_epi16_rmk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsw (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0x08]
+; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rmk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsw (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0x0f]
+; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
+  ret <16 x i16> %3
+}
+
+define <16 x i16> @test_adds_epi16_rmkz_256(<16 x i16> %a, <16 x i16>* %ptr_b, i16 %mask) {
+; X86-LABEL: test_adds_epi16_rmkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsw (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi16_rmkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsw (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i16>, <16 x i16>* %ptr_b
+  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
+  ret <16 x i16> %3
+}
+
+declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) #0
+
+define <16 x i8> @test_adds_epi8_rrk_128(<16 x i8> %a, <16 x i8> %b, <16 x i8> %passThru, i16 %mask) {
+; X86-LABEL: test_adds_epi8_rrk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0xd1]
+; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rrk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0xd1]
+; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_adds_epi8_rrkz_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
+; X86-LABEL: test_adds_epi8_rrkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rrkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_adds_epi8_rm_128(<16 x i8> %a, <16 x i8>* %ptr_b) {
+; X86-LABEL: test_adds_epi8_rm_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vpaddsb (%eax), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rm_128:
+; X64:       # %bb.0:
+; X64-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %1
+}
+
+define <16 x i8> @test_adds_epi8_rmk_128(<16 x i8> %a, <16 x i8>* %ptr_b, <16 x i8> %passThru, i16 %mask) {
+; X86-LABEL: test_adds_epi8_rmk_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsb (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0x08]
+; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rmk_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsb (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0x0f]
+; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
+  ret <16 x i8> %3
+}
+
+define <16 x i8> @test_adds_epi8_rmkz_128(<16 x i8> %a, <16 x i8>* %ptr_b, i16 %mask) {
+; X86-LABEL: test_adds_epi8_rmkz_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsb (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rmkz_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <16 x i8>, <16 x i8>* %ptr_b
+  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
+  %2 = bitcast i16 %mask to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
+  ret <16 x i8> %3
+}
+
+declare <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8>, <32 x i8>) #0
+
+define <32 x i8> @test_adds_epi8_rrk_256(<32 x i8> %a, <32 x i8> %b, <32 x i8> %passThru, i32 %mask) {
+; X86-LABEL: test_adds_epi8_rrk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0xd1]
+; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rrk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0xd1]
+; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_adds_epi8_rrkz_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
+; X86-LABEL: test_adds_epi8_rrkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
+; X86-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rrkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
+; X64-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_adds_epi8_rmk_256(<32 x i8> %a, <32 x i8>* %ptr_b, <32 x i8> %passThru, i32 %mask) {
+; X86-LABEL: test_adds_epi8_rmk_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsb (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0x08]
+; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rmk_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsb (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0x0f]
+; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
+  ret <32 x i8> %3
+}
+
+define <32 x i8> @test_adds_epi8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
+; X86-LABEL: test_adds_epi8_rmkz_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
+; X86-NEXT:    vpaddsb (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0x00]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_adds_epi8_rmkz_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
+; X64-NEXT:    vpaddsb (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0x07]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %b = load <32 x i8>, <32 x i8>* %ptr_b
+  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
+  %2 = bitcast i32 %mask to <32 x i1>
+  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
+  ret <32 x i8> %3
+}
+
 define <16 x i8> @test_mask_adds_epi8_rr_128(<16 x i8> %a, <16 x i8> %b) {
 ; CHECK-LABEL: test_mask_adds_epi8_rr_128:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0xc1]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.avx512.mask.padds.b.128(<16 x i8> %a, <16 x i8> %b, <16 x i8> zeroinitializer, i16 -1)
   ret <16 x i8> %res
 }

 ; X64-NEXT:    vpsubsb (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0x07]
 ; X64-NEXT:    retq # encoding: [0xc3]
   %b = load <32 x i8>, <32 x i8>* %ptr_b
   %res = call <32 x i8> @llvm.x86.avx512.mask.psubs.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.psubs.b.256(<32 x i8>, <32 x i8>, <32 x i8>, i32)
-

test/CodeGen/X86/avx512bwvl-intrinsics.ll

; X64-NEXT:    retq # encoding: [0xc3]
   %1 = call <32 x i8> @llvm.x86.avx2.packuswb(<16 x i16> %a, <16 x i16> %b)
   %2 = bitcast i32 %mask to <32 x i1>
   %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
   ret <32 x i8> %3
 }
 
 declare <32 x i8> @llvm.x86.avx2.packuswb(<16 x i16>, <16 x i16>)
 
-declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) #0
-
-define <8 x i16> @test_mask_adds_epi16_rrk_128(<8 x i16> %a, <8 x i16> %b, <8 x i16> %passThru, i8 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rrk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
-; X86-NEXT:    vpaddsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0xd1]
-; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rrk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0xd1]
-; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
-  ret <8 x i16> %3
-}
-
-define <8 x i16> @test_mask_adds_epi16_rrkz_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rrkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
-; X86-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rrkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
-  ret <8 x i16> %3
-}
-
-define <8 x i16> @test_mask_adds_epi16_rmk_128(<8 x i16> %a, <8 x i16>* %ptr_b, <8 x i16> %passThru, i8 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rmk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
-; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
-; X86-NEXT:    vpaddsw (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0x08]
-; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rmk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsw (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xed,0x0f]
-; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <8 x i16>, <8 x i16>* %ptr_b
-  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
-  ret <8 x i16> %3
-}
-
-define <8 x i16> @test_mask_adds_epi16_rmkz_128(<8 x i16> %a, <8 x i16>* %ptr_b, i8 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rmkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
-; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
-; X86-NEXT:    vpaddsw (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rmkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsw (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xed,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <8 x i16>, <8 x i16>* %ptr_b
-  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
-  ret <8 x i16> %3
-}
-
-declare <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16>, <16 x i16>) #0
-
-define <16 x i16> @test_mask_adds_epi16_rrk_256(<16 x i16> %a, <16 x i16> %b, <16 x i16> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rrk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0xd1]
-; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rrk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0xd1]
-; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
-  ret <16 x i16> %3
-}
-
-define <16 x i16> @test_mask_adds_epi16_rrkz_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rrkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rrkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
-  ret <16 x i16> %3
-}
-
-define <16 x i16> @test_mask_adds_epi16_rmk_256(<16 x i16> %a, <16 x i16>* %ptr_b, <16 x i16> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rmk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsw (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0x08]
-; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rmk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsw (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xed,0x0f]
-; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i16>, <16 x i16>* %ptr_b
-  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
-  ret <16 x i16> %3
-}
-
-define <16 x i16> @test_mask_adds_epi16_rmkz_256(<16 x i16> %a, <16 x i16>* %ptr_b, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi16_rmkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsw (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi16_rmkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsw (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xed,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i16>, <16 x i16>* %ptr_b
-  %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
-  ret <16 x i16> %3
-}
-
-declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) #0
-
-define <8 x i16> @test_mask_subs_epi16_rrk_128(<8 x i16> %a, <8 x i16> %b, <8 x i16> %passThru, i8 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rrk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
-; X86-NEXT:    vpsubsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0xd1]
-; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rrk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsw %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0xd1]
-; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
-  ret <8 x i16> %3
-}
-
-define <8 x i16> @test_mask_subs_epi16_rrkz_128(<8 x i16> %a, <8 x i16> %b, i8 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rrkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
-; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
-; X86-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rrkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
-  ret <8 x i16> %3
-}
-
-define <8 x i16> @test_mask_subs_epi16_rmk_128(<8 x i16> %a, <8 x i16>* %ptr_b, <8 x i16> %passThru, i8 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rmk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
-; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
-; X86-NEXT:    vpsubsw (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0x08]
-; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rmk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsw (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe9,0x0f]
-; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <8 x i16>, <8 x i16>* %ptr_b
-  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> %passThru
-  ret <8 x i16> %3
-}
-
-define <8 x i16> @test_mask_subs_epi16_rmkz_128(<8 x i16> %a, <8 x i16>* %ptr_b, i8 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rmkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
-; X86-NEXT:    kmovd %ecx, %k1 # encoding: [0xc5,0xfb,0x92,0xc9]
-; X86-NEXT:    vpsubsw (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rmkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsw (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe9,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <8 x i16>, <8 x i16>* %ptr_b
-  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a, <8 x i16> %b)
-  %2 = bitcast i8 %mask to <8 x i1>
-  %3 = select <8 x i1> %2, <8 x i16> %1, <8 x i16> zeroinitializer
-  ret <8 x i16> %3
-}
-
-declare <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16>, <16 x i16>) #0
-
-define <16 x i16> @test_mask_subs_epi16_rrk_256(<16 x i16> %a, <16 x i16> %b, <16 x i16> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rrk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0xd1]
-; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rrk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsw %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0xd1]
-; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
-  ret <16 x i16> %3
-}
-
-define <16 x i16> @test_mask_subs_epi16_rrkz_256(<16 x i16> %a, <16 x i16> %b, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rrkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rrkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsw %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
-  ret <16 x i16> %3
-}
-
-define <16 x i16> @test_mask_subs_epi16_rmk_256(<16 x i16> %a, <16 x i16>* %ptr_b, <16 x i16> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rmk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsw (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0x08]
-; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rmk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsw (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe9,0x0f]
-; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i16>, <16 x i16>* %ptr_b
-  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> %passThru
-  ret <16 x i16> %3
-}
-
-define <16 x i16> @test_mask_subs_epi16_rmkz_256(<16 x i16> %a, <16 x i16>* %ptr_b, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi16_rmkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsw (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi16_rmkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsw (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe9,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i16>, <16 x i16>* %ptr_b
-  %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> %a, <16 x i16> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i16> %1, <16 x i16> zeroinitializer
-  ret <16 x i16> %3
-}
-
-declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) #0
-
-define <16 x i8> @test_mask_adds_epi8_rrk_128(<16 x i8> %a, <16 x i8> %b, <16 x i8> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rrk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0xd1]
-; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rrk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0xd1]
-; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
-  ret <16 x i8> %3
-}
-
-define <16 x i8> @test_mask_adds_epi8_rrkz_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rrkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rrkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
-  ret <16 x i8> %3
-}
-
-define <16 x i8> @test_mask_adds_epi8_rm_128(<16 x i8> %a, <16 x i8>* %ptr_b) {
-; X86-LABEL: test_mask_adds_epi8_rm_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    vpaddsb (%eax), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rm_128:
-; X64:       # %bb.0:
-; X64-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i8>, <16 x i8>* %ptr_b
-  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
-  ret <16 x i8> %1
-}
-
-define <16 x i8> @test_mask_adds_epi8_rmk_128(<16 x i8> %a, <16 x i8>* %ptr_b, <16 x i8> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rmk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsb (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0x08]
-; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rmk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsb (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xec,0x0f]
-; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i8>, <16 x i8>* %ptr_b
-  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
-  ret <16 x i8> %3
-}
-
-define <16 x i8> @test_mask_adds_epi8_rmkz_128(<16 x i8> %a, <16 x i8>* %ptr_b, i16 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rmkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsb (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rmkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsb (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xec,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i8>, <16 x i8>* %ptr_b
-  %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
-  ret <16 x i8> %3
-}
-
-declare <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8>, <32 x i8>) #0
-
-define <32 x i8> @test_mask_adds_epi8_rrk_256(<32 x i8> %a, <32 x i8> %b, <32 x i8> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rrk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0xd1]
-; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rrk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0xd1]
-; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
-  ret <32 x i8> %3
-}
-
-define <32 x i8> @test_mask_adds_epi8_rrkz_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rrkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rrkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpaddsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
-  ret <32 x i8> %3
-}
-
-define <32 x i8> @test_mask_adds_epi8_rmk_256(<32 x i8> %a, <32 x i8>* %ptr_b, <32 x i8> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rmk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsb (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0x08]
-; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rmk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsb (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xec,0x0f]
-; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i8>, <32 x i8>* %ptr_b
-  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
-  ret <32 x i8> %3
-}
-
-define <32 x i8> @test_mask_adds_epi8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
-; X86-LABEL: test_mask_adds_epi8_rmkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpaddsb (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_adds_epi8_rmkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpaddsb (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xec,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i8>, <32 x i8>* %ptr_b
-  %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
-  ret <32 x i8> %3
-}
-
-declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) #0
-
-define <16 x i8> @test_mask_subs_epi8_rrk_128(<16 x i8> %a, <16 x i8> %b, <16 x i8> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rrk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0xd1]
-; X86-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rrk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsb %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0xd1]
-; X64-NEXT:    vmovdqa %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
-  ret <16 x i8> %3
-}
-
-define <16 x i8> @test_mask_subs_epi8_rrkz_128(<16 x i8> %a, <16 x i8> %b, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rrkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rrkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
-  ret <16 x i8> %3
-}
-
-define <16 x i8> @test_mask_subs_epi8_rmk_128(<16 x i8> %a, <16 x i8>* %ptr_b, <16 x i8> %passThru, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rmk_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsb (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0x08]
-; X86-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rmk_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsb (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x09,0xe8,0x0f]
-; X64-NEXT:    vmovdqa %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i8>, <16 x i8>* %ptr_b
-  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> %passThru
-  ret <16 x i8> %3
-}
-
-define <16 x i8> @test_mask_subs_epi8_rmkz_128(<16 x i8> %a, <16 x i8>* %ptr_b, i16 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rmkz_128:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsb (%eax), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rmkz_128:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsb (%rdi), %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0x89,0xe8,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <16 x i8>, <16 x i8>* %ptr_b
-  %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a, <16 x i8> %b)
-  %2 = bitcast i16 %mask to <16 x i1>
-  %3 = select <16 x i1> %2, <16 x i8> %1, <16 x i8> zeroinitializer
-  ret <16 x i8> %3
-}
-
-declare <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8>, <32 x i8>) #0
-
-define <32 x i8> @test_mask_subs_epi8_rrk_256(<32 x i8> %a, <32 x i8> %b, <32 x i8> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rrk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0xd1]
-; X86-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rrk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsb %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0xd1]
-; X64-NEXT:    vmovdqa %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc2]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
-  ret <32 x i8> %3
-}
-
-define <32 x i8> @test_mask_subs_epi8_rrkz_256(<32 x i8> %a, <32 x i8> %b, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rrkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
-; X86-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rrkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
-; X64-NEXT:    vpsubsb %ymm1, %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
-  ret <32 x i8> %3
-}
-
-define <32 x i8> @test_mask_subs_epi8_rmk_256(<32 x i8> %a, <32 x i8>* %ptr_b, <32 x i8> %passThru, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rmk_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsb (%eax), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0x08]
-; X86-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rmk_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsb (%rdi), %ymm0, %ymm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x29,0xe8,0x0f]
-; X64-NEXT:    vmovdqa %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x6f,0xc1]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i8>, <32 x i8>* %ptr_b
-  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> %passThru
-  ret <32 x i8> %3
-}
-
-define <32 x i8> @test_mask_subs_epi8_rmkz_256(<32 x i8> %a, <32 x i8>* %ptr_b, i32 %mask) {
-; X86-LABEL: test_mask_subs_epi8_rmkz_256:
-; X86:       # %bb.0:
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
-; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
-; X86-NEXT:    vpsubsb (%eax), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0x00]
-; X86-NEXT:    retl # encoding: [0xc3]
-;
-; X64-LABEL: test_mask_subs_epi8_rmkz_256:
-; X64:       # %bb.0:
-; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
-; X64-NEXT:    vpsubsb (%rdi), %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xa9,0xe8,0x07]
-; X64-NEXT:    retq # encoding: [0xc3]
-  %b = load <32 x i8>, <32 x i8>* %ptr_b
-  %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> %a, <32 x i8> %b)
-  %2 = bitcast i32 %mask to <32 x i1>
-  %3 = select <32 x i1> %2, <32 x i8> %1, <32 x i8> zeroinitializer
-  ret <32 x i8> %3
-}
-
 define <8 x i16>@test_int_x86_avx512_mask_vpermt2var_hi_128(<8 x i16> %x0, <8 x i16> %x1, <8 x i16> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask_vpermt2var_hi_128:
 ; X86:       # %bb.0:
 ; X86-NEXT:    vmovdqa %xmm1, %xmm3 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x6f,0xd9]
 ; X86-NEXT:    vpermt2w %xmm2, %xmm0, %xmm3 # encoding: [0x62,0xf2,0xfd,0x08,0x7d,0xda]
 ; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
 ; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
 ; X86-NEXT:    vpermt2w %xmm2, %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0x7d,0xca]

test/CodeGen/X86/pic-load-remat.ll

 ; RUN: llc < %s -mtriple=i686-apple-darwin -mattr=+sse2 -relocation-model=pic | grep psllw | grep pb
 
 define void @f() nounwind  {
 entry:
   br label %bb
 
 bb:   ; preds = %bb, %entry
-  %tmp4403 = tail call <8 x i16> @llvm.x86.sse2.psubs.w( <8 x i16> zeroinitializer, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=2]
-  %tmp4443 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> zeroinitializer, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4403 = tail call <8 x i16> @llvm.ssub.sat.v8i16( <8 x i16> zeroinitializer, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=2]
+  %tmp4443 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> zeroinitializer, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4609 = tail call <8 x i16> @llvm.x86.sse2.psll.w( <8 x i16> zeroinitializer, <8 x i16> bitcast (<4 x i32> < i32 3, i32 5, i32 6, i32 9 > to <8 x i16>) )    ; <<8 x i16>> [#uses=1]
   %tmp4651 = add <8 x i16> %tmp4609, < i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1 >    ; <<8 x i16>> [#uses=1]
   %tmp4658 = tail call <8 x i16> @llvm.x86.sse2.psll.w( <8 x i16> %tmp4651, <8 x i16> bitcast (<4 x i32> < i32 4, i32 1, i32 2, i32 3 > to <8 x i16>) )    ; <<8 x i16>> [#uses=1]
   %tmp4669 = tail call <8 x i16> @llvm.x86.sse2.pavg.w( <8 x i16> < i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170 >, <8 x i16> %tmp4443 ) nounwind readnone     ; <<8 x i16>> [#uses=2]
-  %tmp4679 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> %tmp4669, <8 x i16> %tmp4669 ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4679 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> %tmp4669, <8 x i16> %tmp4669 ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4689 = add <8 x i16> %tmp4679, %tmp4658    ; <<8 x i16>> [#uses=1]
-  %tmp4700 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> %tmp4689, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4700 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> %tmp4689, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4708 = bitcast <8 x i16> %tmp4700 to <2 x i64>    ; <<2 x i64>> [#uses=1]
   %tmp4772 = add <8 x i16> zeroinitializer, < i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1 >    ; <<8 x i16>> [#uses=1]
   %tmp4779 = tail call <8 x i16> @llvm.x86.sse2.psll.w( <8 x i16> %tmp4772, <8 x i16> bitcast (<4 x i32> < i32 3, i32 5, i32 undef, i32 7 > to <8 x i16>) )    ; <<8 x i16>> [#uses=1]
   %tmp4810 = add <8 x i16> zeroinitializer, %tmp4779    ; <<8 x i16>> [#uses=1]
-  %tmp4821 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> %tmp4810, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4821 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> %tmp4810, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4829 = bitcast <8 x i16> %tmp4821 to <2 x i64>    ; <<2 x i64>> [#uses=1]
   %tmp4900 = tail call <8 x i16> @llvm.x86.sse2.psll.w( <8 x i16> zeroinitializer, <8 x i16> bitcast (<4 x i32> < i32 1, i32 1, i32 2, i32 2 > to <8 x i16>) )    ; <<8 x i16>> [#uses=1]
   %tmp4911 = tail call <8 x i16> @llvm.x86.sse2.pavg.w( <8 x i16> < i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170, i16 -23170 >, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=2]
-  %tmp4921 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> %tmp4911, <8 x i16> %tmp4911 ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4921 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> %tmp4911, <8 x i16> %tmp4911 ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4931 = add <8 x i16> %tmp4921, %tmp4900    ; <<8 x i16>> [#uses=1]
-  %tmp4942 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> %tmp4931, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4942 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> %tmp4931, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4950 = bitcast <8 x i16> %tmp4942 to <2 x i64>    ; <<2 x i64>> [#uses=1]
-  %tmp4957 = tail call <8 x i16> @llvm.x86.sse2.padds.w( <8 x i16> %tmp4403, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4957 = tail call <8 x i16> @llvm.sadd.sat.v8i16( <8 x i16> %tmp4403, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4958 = bitcast <8 x i16> %tmp4957 to <2 x i64>    ; <<2 x i64>> [#uses=1]
-  %tmp4967 = tail call <8 x i16> @llvm.x86.sse2.psubs.w( <8 x i16> %tmp4403, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
+  %tmp4967 = tail call <8 x i16> @llvm.ssub.sat.v8i16( <8 x i16> %tmp4403, <8 x i16> zeroinitializer ) nounwind readnone     ; <<8 x i16>> [#uses=1]
   %tmp4968 = bitcast <8 x i16> %tmp4967 to <2 x i64>    ; <<2 x i64>> [#uses=1]
   store <2 x i64> %tmp4829, <2 x i64>* null, align 16
   store <2 x i64> %tmp4958, <2 x i64>* null, align 16
   store <2 x i64> %tmp4968, <2 x i64>* null, align 16
   store <2 x i64> %tmp4950, <2 x i64>* null, align 16
   store <2 x i64> %tmp4708, <2 x i64>* null, align 16
   br label %bb
 }
 
 declare <8 x i16> @llvm.x86.sse2.psll.w(<8 x i16>, <8 x i16>) nounwind readnone 
 
 declare <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16>, <8 x i16>) nounwind readnone 
 
-declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone 
+declare <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16>, <8 x i16>) nounwind readnone 
 
-declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone 
+declare <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16>, <8 x i16>) nounwind readnone 

test/CodeGen/X86/sse2-intrinsics-fast-isel.ll

 ; AVX1-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX512-LABEL: test_mm_adds_epi8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0xc1]
 ; AVX512-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %arg0 = bitcast <2 x i64> %a0 to <16 x i8>
   %arg1 = bitcast <2 x i64> %a1 to <16 x i8>
-  %res = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %arg0, <16 x i8> %arg1)
+  %res = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %arg0, <16 x i8> %arg1)
   %bc = bitcast <16 x i8> %res to <2 x i64>
   ret <2 x i64> %bc
 }
-declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone
+declare <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
 
 define <2 x i64> @test_mm_adds_epi16(<2 x i64> %a0, <2 x i64> %a1) nounwind {
 ; SSE-LABEL: test_mm_adds_epi16:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    paddsw %xmm1, %xmm0 # encoding: [0x66,0x0f,0xed,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX1-LABEL: test_mm_adds_epi16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xed,0xc1]
 ; AVX1-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX512-LABEL: test_mm_adds_epi16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xed,0xc1]
 ; AVX512-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
   %arg1 = bitcast <2 x i64> %a1 to <8 x i16>
-  %res = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %arg0, <8 x i16> %arg1)
+  %res = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %arg0, <8 x i16> %arg1)
   %bc = bitcast <8 x i16> %res to <2 x i64>
   ret <2 x i64> %bc
 }
-declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone
+declare <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
 
 define <2 x i64> @test_mm_adds_epu8(<2 x i64> %a0, <2 x i64> %a1) nounwind {
 ; SSE-LABEL: test_mm_adds_epu8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    paddusb %xmm1, %xmm0 # encoding: [0x66,0x0f,0xdc,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX1-LABEL: test_mm_adds_epu8:

 ; AVX1-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX512-LABEL: test_mm_subs_epi8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe8,0xc1]
 ; AVX512-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %arg0 = bitcast <2 x i64> %a0 to <16 x i8>
   %arg1 = bitcast <2 x i64> %a1 to <16 x i8>
-  %res = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %arg0, <16 x i8> %arg1)
+  %res = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %arg0, <16 x i8> %arg1)
   %bc = bitcast <16 x i8> %res to <2 x i64>
   ret <2 x i64> %bc
 }
-declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone
+declare <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
 
 define <2 x i64> @test_mm_subs_epi16(<2 x i64> %a0, <2 x i64> %a1) nounwind {
 ; SSE-LABEL: test_mm_subs_epi16:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    psubsw %xmm1, %xmm0 # encoding: [0x66,0x0f,0xe9,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX1-LABEL: test_mm_subs_epi16:
 ; AVX1:       # %bb.0:
 ; AVX1-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 # encoding: [0xc5,0xf9,0xe9,0xc1]
 ; AVX1-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX512-LABEL: test_mm_subs_epi16:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe9,0xc1]
 ; AVX512-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
   %arg0 = bitcast <2 x i64> %a0 to <8 x i16>
   %arg1 = bitcast <2 x i64> %a1 to <8 x i16>
-  %res = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %arg0, <8 x i16> %arg1)
+  %res = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %arg0, <8 x i16> %arg1)
   %bc = bitcast <8 x i16> %res to <2 x i64>
   ret <2 x i64> %bc
 }
-declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
+declare <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
 
 define <2 x i64> @test_mm_subs_epu8(<2 x i64> %a0, <2 x i64> %a1) nounwind {
 ; SSE-LABEL: test_mm_subs_epu8:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    psubusb %xmm1, %xmm0 # encoding: [0x66,0x0f,0xd8,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
 ;
 ; AVX1-LABEL: test_mm_subs_epu8:

test/CodeGen/X86/sse2-intrinsics-x86-upgrade.ll

 ; AVX512-NEXT:    vcvtdq2ps %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x5b,0xc0]
 ; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %res = call <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32> %a0) ; <<4 x float>> [#uses=1]
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.sse2.cvtdq2ps(<4 x i32>) nounwind readnone
 
 
+define <16 x i8> @test_x86_sse2_padds_b(<16 x i8> %a0, <16 x i8> %a1) {
+; SSE-LABEL: test_x86_sse2_padds_b:
+; SSE:       ## %bb.0:
+; SSE-NEXT:    paddsb %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xec,0xc1]
+; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX1-LABEL: test_x86_sse2_padds_b:
+; AVX1:       ## %bb.0:
+; AVX1-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xec,0xc1]
+; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX512-LABEL: test_x86_sse2_padds_b:
+; AVX512:       ## %bb.0:
+; AVX512-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0xc1]
+; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+  %res = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
+  ret <16 x i8> %res
+}
+declare <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+
+
+define <8 x i16> @test_x86_sse2_padds_w(<8 x i16> %a0, <8 x i16> %a1) {
+; SSE-LABEL: test_x86_sse2_padds_w:
+; SSE:       ## %bb.0:
+; SSE-NEXT:    paddsw %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xed,0xc1]
+; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX1-LABEL: test_x86_sse2_padds_w:
+; AVX1:       ## %bb.0:
+; AVX1-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xed,0xc1]
+; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX512-LABEL: test_x86_sse2_padds_w:
+; AVX512:       ## %bb.0:
+; AVX512-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xed,0xc1]
+; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+  %res = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a0, <8 x i16> %a1) ; <<8 x i16>> [#uses=1]
+  ret <8 x i16> %res
+}
+declare <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+
+
 define <16 x i8> @test_x86_sse2_paddus_b(<16 x i8> %a0, <16 x i8> %a1) {
 ; SSE-LABEL: test_x86_sse2_paddus_b:
 ; SSE:       ## %bb.0:
 ; SSE-NEXT:    paddusb %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xdc,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
 ;
 ; AVX2-LABEL: test_x86_sse2_paddus_b:
 ; AVX2:       ## %bb.0:

 ; SKX-NEXT:    vpaddusw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xdd,0xc1]
 ; SKX-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %res = call <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16> %a0, <8 x i16> %a1) ; <<8 x i16>> [#uses=1]
   ret <8 x i16> %res
 }
 declare <8 x i16> @llvm.x86.sse2.paddus.w(<8 x i16>, <8 x i16>) nounwind readnone
 
 
+define <16 x i8> @test_x86_sse2_psubs_b(<16 x i8> %a0, <16 x i8> %a1) {
+; SSE-LABEL: test_x86_sse2_psubs_b:
+; SSE:       ## %bb.0:
+; SSE-NEXT:    psubsb %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xe8,0xc1]
+; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX1-LABEL: test_x86_sse2_psubs_b:
+; AVX1:       ## %bb.0:
+; AVX1-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xe8,0xc1]
+; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX512-LABEL: test_x86_sse2_psubs_b:
+; AVX512:       ## %bb.0:
+; AVX512-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe8,0xc1]
+; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+  %res = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
+  ret <16 x i8> %res
+}
+declare <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
+
+
+define <8 x i16> @test_x86_sse2_psubs_w(<8 x i16> %a0, <8 x i16> %a1) {
+; SSE-LABEL: test_x86_sse2_psubs_w:
+; SSE:       ## %bb.0:
+; SSE-NEXT:    psubsw %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xe9,0xc1]
+; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX1-LABEL: test_x86_sse2_psubs_w:
+; AVX1:       ## %bb.0:
+; AVX1-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xe9,0xc1]
+; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+;
+; AVX512-LABEL: test_x86_sse2_psubs_w:
+; AVX512:       ## %bb.0:
+; AVX512-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe9,0xc1]
+; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
+  %res = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a0, <8 x i16> %a1) ; <<8 x i16>> [#uses=1]
+  ret <8 x i16> %res
+}
+declare <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16>, <8 x i16>) nounwind readnone
+
+
 define <16 x i8> @test_x86_sse2_psubus_b(<16 x i8> %a0, <16 x i8> %a1) {
 ; SSE-LABEL: test_x86_sse2_psubus_b:
 ; SSE:       ## %bb.0:
 ; SSE-NEXT:    psubusb %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xd8,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
 ;
 ; AVX2-LABEL: test_x86_sse2_psubus_b:
 ; AVX2:       ## %bb.0:

test/CodeGen/X86/sse2-intrinsics-x86.ll

 ; X64-AVX512-NEXT:    ## encoding: [0xc5,0xf8,0x28,0x05,A,A,A,A]
 ; X64-AVX512-NEXT:    ## fixup A - offset: 4, value: LCPI34_0-4, kind: reloc_riprel_4byte
 ; X64-AVX512-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.sse2.packuswb.128(<8 x i16> <i16 0, i16 255, i16 256, i16 65535, i16 -1, i16 -255, i16 -256, i16 -32678>, <8 x i16> zeroinitializer)
   ret <16 x i8> %res
 }
 
 
-define <16 x i8> @test_x86_sse2_padds_b(<16 x i8> %a0, <16 x i8> %a1) {
-; SSE-LABEL: test_x86_sse2_padds_b:
-; SSE:       ## %bb.0:
-; SSE-NEXT:    paddsb %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xec,0xc1]
-; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX1-LABEL: test_x86_sse2_padds_b:
-; AVX1:       ## %bb.0:
-; AVX1-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xec,0xc1]
-; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512-LABEL: test_x86_sse2_padds_b:
-; AVX512:       ## %bb.0:
-; AVX512-NEXT:    vpaddsb %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xec,0xc1]
-; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
-  ret <16 x i8> %res
-}
-declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone
-
-
-define <8 x i16> @test_x86_sse2_padds_w(<8 x i16> %a0, <8 x i16> %a1) {
-; SSE-LABEL: test_x86_sse2_padds_w:
-; SSE:       ## %bb.0:
-; SSE-NEXT:    paddsw %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xed,0xc1]
-; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX1-LABEL: test_x86_sse2_padds_w:
-; AVX1:       ## %bb.0:
-; AVX1-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xed,0xc1]
-; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512-LABEL: test_x86_sse2_padds_w:
-; AVX512:       ## %bb.0:
-; AVX512-NEXT:    vpaddsw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xed,0xc1]
-; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a0, <8 x i16> %a1) ; <<8 x i16>> [#uses=1]
-  ret <8 x i16> %res
-}
-declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-
 define <4 x i32> @test_x86_sse2_pmadd_wd(<8 x i16> %a0, <8 x i16> %a1) {
 ; SSE-LABEL: test_x86_sse2_pmadd_wd:
 ; SSE:       ## %bb.0:
 ; SSE-NEXT:    pmaddwd %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xf5,0xc1]
 ; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
 ;
 ; AVX1-LABEL: test_x86_sse2_pmadd_wd:
 ; AVX1:       ## %bb.0:

 ; AVX512-NEXT:    vpsrlw $7, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0x71,0xd0,0x07]
 ; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
   %res = call <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16> %a0, i32 7) ; <<8 x i16>> [#uses=1]
   ret <8 x i16> %res
 }
 declare <8 x i16> @llvm.x86.sse2.psrli.w(<8 x i16>, i32) nounwind readnone
 
 
-define <16 x i8> @test_x86_sse2_psubs_b(<16 x i8> %a0, <16 x i8> %a1) {
-; SSE-LABEL: test_x86_sse2_psubs_b:
-; SSE:       ## %bb.0:
-; SSE-NEXT:    psubsb %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xe8,0xc1]
-; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX1-LABEL: test_x86_sse2_psubs_b:
-; AVX1:       ## %bb.0:
-; AVX1-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xe8,0xc1]
-; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512-LABEL: test_x86_sse2_psubs_b:
-; AVX512:       ## %bb.0:
-; AVX512-NEXT:    vpsubsb %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe8,0xc1]
-; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> %a0, <16 x i8> %a1) ; <<16 x i8>> [#uses=1]
-  ret <16 x i8> %res
-}
-declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone
-
-
-define <8 x i16> @test_x86_sse2_psubs_w(<8 x i16> %a0, <8 x i16> %a1) {
-; SSE-LABEL: test_x86_sse2_psubs_w:
-; SSE:       ## %bb.0:
-; SSE-NEXT:    psubsw %xmm1, %xmm0 ## encoding: [0x66,0x0f,0xe9,0xc1]
-; SSE-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX1-LABEL: test_x86_sse2_psubs_w:
-; AVX1:       ## %bb.0:
-; AVX1-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 ## encoding: [0xc5,0xf9,0xe9,0xc1]
-; AVX1-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-;
-; AVX512-LABEL: test_x86_sse2_psubs_w:
-; AVX512:       ## %bb.0:
-; AVX512-NEXT:    vpsubsw %xmm1, %xmm0, %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xf9,0xe9,0xc1]
-; AVX512-NEXT:    ret{{[l|q]}} ## encoding: [0xc3]
-  %res = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> %a0, <8 x i16> %a1) ; <<8 x i16>> [#uses=1]
-  ret <8 x i16> %res
-}
-declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
-
-
 define i32 @test_x86_sse2_ucomieq_sd(<2 x double> %a0, <2 x double> %a1) {
 ; SSE-LABEL: test_x86_sse2_ucomieq_sd:
 ; SSE:       ## %bb.0:
 ; SSE-NEXT:    ucomisd %xmm1, %xmm0 ## encoding: [0x66,0x0f,0x2e,0xc1]
 ; SSE-NEXT:    setnp %al ## encoding: [0x0f,0x9b,0xc0]
 ; SSE-NEXT:    sete %cl ## encoding: [0x0f,0x94,0xc1]
 ; SSE-NEXT:    andb %al, %cl ## encoding: [0x20,0xc1]
 ; SSE-NEXT:    movzbl %cl, %eax ## encoding: [0x0f,0xb6,0xc1]

test/Instrumentation/MemorySanitizer/msan_x86intrinsics.ll

 ; CHECK-ORIGINS: store i32 {{.*}}[[ORIGIN]], i32* @__msan_retval_origin_tls
 ; CHECK: ret <16 x i8>
 
 
 ; Simple NoMem intrinsic
 ; Check that shadow is OR'ed, and origin is Select'ed
 ; And no shadow checks!
 
-define <8 x i16> @Paddsw128(<8 x i16> %a, <8 x i16> %b) nounwind uwtable sanitize_memory {
-  %call = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b)
+define <8 x i16> @Pmulhuw128(<8 x i16> %a, <8 x i16> %b) nounwind uwtable sanitize_memory {
+  %call = call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> %a, <8 x i16> %b)
   ret <8 x i16> %call
 }
 
-declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> %a, <8 x i16> %b) nounwind
+declare <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> %a, <8 x i16> %b) nounwind
 
-; CHECK-LABEL: @Paddsw128
+; CHECK-LABEL: @Pmulhuw128
 ; CHECK-NEXT: load <8 x i16>, <8 x i16>* {{.*}} @__msan_param_tls
 ; CHECK-ORIGINS: load i32, i32* {{.*}} @__msan_param_origin_tls
 ; CHECK-NEXT: load <8 x i16>, <8 x i16>* {{.*}} @__msan_param_tls
 ; CHECK-ORIGINS: load i32, i32* {{.*}} @__msan_param_origin_tls
 ; CHECK-NEXT: = or <8 x i16>
 ; CHECK-ORIGINS: = bitcast <8 x i16> {{.*}} to i128
 ; CHECK-ORIGINS-NEXT: = icmp ne i128 {{.*}}, 0
 ; CHECK-ORIGINS-NEXT: = select i1 {{.*}}, i32 {{.*}}, i32
-; CHECK-NEXT: call <8 x i16> @llvm.x86.sse2.padds.w
+; CHECK-NEXT: call <8 x i16> @llvm.x86.sse2.pmulhu.w
 ; CHECK-NEXT: store <8 x i16> {{.*}} @__msan_retval_tls
 ; CHECK-ORIGINS: store i32 {{.*}} @__msan_retval_origin_tls
 ; CHECK-NEXT: ret <8 x i16>

test/Transforms/InstCombine/X86/x86-adds-subs.ll

 ; CHECK-NEXT:    ret <16 x i8> <i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 -128, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33>
 ;
   %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 -107, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 -48, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
   ret <16 x i8> %1
 }
 
 define <16 x i8> @sse2_adds_b_constant_overflow() {
 ; CHECK-LABEL: @sse2_adds_b_constant_overflow(
 ; CHECK-NEXT:    ret <16 x i8> <i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127>
 ;
   %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16>, <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120>)
   ret <16 x i8> %1
 }
 
 define <16 x i8> @sse2_adds_b_constant_undefs() {
 ; CHECK-LABEL: @sse2_adds_b_constant_undefs(
-; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <16 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
+; CHECK-NEXT:    ret <16 x i8> [[TMP1]]
 ;
   %1 = call <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <16 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
   ret <16 x i8> %1
 }
 
 define <32 x i8> @avx2_adds_b_constant() {
 ; CHECK-LABEL: @avx2_adds_b_constant(
 ; CHECK-NEXT:    ret <32 x i8> <i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33>

 ; CHECK-NEXT:    ret <32 x i8> <i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127>
 ;
   %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16>, <32 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120>)
   ret <32 x i8> %1
 }
 
 define <32 x i8> @avx2_adds_b_constant_undefs() {
 ; CHECK-LABEL: @avx2_adds_b_constant_undefs(
-; CHECK-NEXT:    ret <32 x i8> <i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <32 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
+; CHECK-NEXT:    ret <32 x i8> [[TMP1]]
 ;
   %1 = call <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <32 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
   ret <32 x i8> %1
 }
 
 define <64 x i8> @avx512_mask_adds_b_constant() {
 ; CHECK-LABEL: @avx512_mask_adds_b_constant(
 ; CHECK-NEXT:    ret <64 x i8> <i8 2, i8 0, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33>

 ; CHECK-NEXT:    ret <64 x i8> <i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127, i8 2, i8 4, i8 6, i8 8, i8 10, i8 12, i8 14, i8 16, i8 18, i8 20, i8 127, i8 24, i8 26, i8 28, i8 30, i8 127>
 ;
   %1 = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 125, i8 12, i8 13, i8 14, i8 15, i8 16>, <64 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 120>, <64 x i8> zeroinitializer, i64 -1)
   ret <64 x i8> %1
 }
 
 define <64 x i8> @avx512_mask_adds_b_constant_undefs() {
 ; CHECK-LABEL: @avx512_mask_adds_b_constant_undefs(
-; CHECK-NEXT:    ret <64 x i8> <i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33, i8 undef, i8 4, i8 6, i8 8, i8 10, i8 undef, i8 14, i8 16, i8 18, i8 20, i8 22, i8 24, i8 26, i8 28, i8 30, i8 33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <64 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
+; CHECK-NEXT:    ret <64 x i8> [[TMP1]]
 ;
   %1 = call <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 undef, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16>, <64 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>, <64 x i8> zeroinitializer, i64 -1)
   ret <64 x i8> %1
 }
 
 define <8 x i16> @sse2_adds_w_constant() {
 ; CHECK-LABEL: @sse2_adds_w_constant(
 ; CHECK-NEXT:    ret <8 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16>

 ; CHECK-NEXT:    ret <8 x i16> <i16 2, i16 32767, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16>
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> <i16 1, i16 8248, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>, <8 x i16> <i16 1, i16 25192, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
   ret <8 x i16> %1
 }
 
 define <8 x i16> @sse2_adds_w_constant_undefs() {
 ; CHECK-LABEL: @sse2_adds_w_constant_undefs(
-; CHECK-NEXT:    ret <8 x i16> <i16 undef, i16 4, i16 6, i16 8, i16 10, i16 undef, i16 14, i16 16>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8>, <8 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
+; CHECK-NEXT:    ret <8 x i16> [[TMP1]]
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8>, <8 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
   ret <8 x i16> %1
 }
 
 define <16 x i16> @avx2_adds_w_constant() {
 ; CHECK-LABEL: @avx2_adds_w_constant(
 ; CHECK-NEXT:    ret <16 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33>

 ; CHECK-NEXT:    ret <16 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16, i16 18, i16 20, i16 32767, i16 24, i16 26, i16 28, i16 30, i16 32767>
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 20125, i16 12, i16 13, i16 14, i16 15, i16 20160>, <16 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 20230, i16 12, i16 13, i16 14, i16 15, i16 20120>)
   ret <16 x i16> %1
 }
 
 define <16 x i16> @avx2_adds_w_constant_undefs() {
 ; CHECK-LABEL: @avx2_adds_w_constant_undefs(
-; CHECK-NEXT:    ret <16 x i16> <i16 undef, i16 4, i16 6, i16 8, i16 10, i16 undef, i16 14, i16 16, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <16 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
+; CHECK-NEXT:    ret <16 x i16> [[TMP1]]
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <16 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
   ret <16 x i16> %1
 }
 
 define <32 x i16> @avx512_mask_adds_w_constant() {
 ; CHECK-LABEL: @avx512_mask_adds_w_constant(
 ; CHECK-NEXT:    ret <32 x i16> <i16 2, i16 0, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33, i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33>

 ; CHECK-NEXT:    ret <32 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16, i16 18, i16 20, i16 32767, i16 24, i16 26, i16 28, i16 30, i16 32767, i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 16, i16 18, i16 20, i16 32767, i16 24, i16 26, i16 28, i16 30, i16 32767>
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 20125, i16 12, i16 13, i16 14, i16 15, i16 20200, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 20125, i16 12, i16 13, i16 14, i16 15, i16 20200>, <32 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 20200, i16 12, i16 13, i16 14, i16 15, i16 20120, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 20211, i16 12, i16 13, i16 14, i16 15, i16 20120>, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %1
 }
 
 define <32 x i16> @avx512_mask_adds_w_constant_undefs() {
 ; CHECK-LABEL: @avx512_mask_adds_w_constant_undefs(
-; CHECK-NEXT:    ret <32 x i16> <i16 undef, i16 4, i16 6, i16 8, i16 10, i16 undef, i16 14, i16 16, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33, i16 undef, i16 4, i16 6, i16 8, i16 10, i16 undef, i16 14, i16 16, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16, i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <32 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
+; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16, i16 undef, i16 2, i16 3, i16 4, i16 5, i16 undef, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <32 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %1
 }
 
 define <16 x i8> @sse2_subs_b_constant() {
 ; CHECK-LABEL: @sse2_subs_b_constant(
 ; CHECK-NEXT:    ret <16 x i8> <i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33>

 ; CHECK-NEXT:    ret <16 x i8> <i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127>
 ;
   %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> <i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16>, <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120>)
   ret <16 x i8> %1
 }
 
 define <16 x i8> @sse2_subs_b_constant_undefs() {
 ; CHECK-LABEL: @sse2_subs_b_constant_undefs(
-; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> <i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16>, <16 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
+; CHECK-NEXT:    ret <16 x i8> [[TMP1]]
 ;
   %1 = call <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8> <i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16>, <16 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
   ret <16 x i8> %1
 }
 
 define <32 x i8> @avx2_subs_b_constant() {
 ; CHECK-LABEL: @avx2_subs_b_constant(
 ; CHECK-NEXT:    ret <32 x i8> <i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33>

 ; CHECK-NEXT:    ret <32 x i8> <i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127>
 ;
   %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> <i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16, i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16>, <32 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120>)
   ret <32 x i8> %1
 }
 
 define <32 x i8> @avx2_subs_b_constant_undefs() {
 ; CHECK-LABEL: @avx2_subs_b_constant_undefs(
-; CHECK-NEXT:    ret <32 x i8> <i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> <i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16>, <32 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
+; CHECK-NEXT:    ret <32 x i8> [[TMP1]]
 ;
   %1 = call <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8> <i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16>, <32 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
   ret <32 x i8> %1
 }
 
 define <64 x i8> @avx512_mask_subs_b_constant() {
 ; CHECK-LABEL: @avx512_mask_subs_b_constant(
 ; CHECK-NEXT:    ret <64 x i8> <i8 -2, i8 0, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33>

 ; CHECK-NEXT:    ret <64 x i8> <i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127, i8 -2, i8 -4, i8 -6, i8 -8, i8 -10, i8 -12, i8 -14, i8 -16, i8 -18, i8 -20, i8 127, i8 -24, i8 -26, i8 -28, i8 -30, i8 127>
 ;
   %1 = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> <i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16, i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16, i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16, i8 -1, i8 -2, i8 -3, i8 -4, i8 -5, i8 -6, i8 -7, i8 -8, i8 -9, i8 -10, i8 125, i8 -12, i8 -13, i8 -14, i8 -15, i8 16>, <64 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 -11, i8 12, i8 13, i8 14, i8 15, i8 -120>, <64 x i8> zeroinitializer, i64 -1)
   ret <64 x i8> %1
 }
 
 define <64 x i8> @avx512_mask_subs_b_constant_undefs() {
 ; CHECK-LABEL: @avx512_mask_subs_b_constant_undefs(
-; CHECK-NEXT:    ret <64 x i8> <i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33, i8 undef, i8 -4, i8 -6, i8 -8, i8 -10, i8 undef, i8 -14, i8 -16, i8 -18, i8 -20, i8 -22, i8 -24, i8 -26, i8 -28, i8 -30, i8 -33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> <i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16>, <64 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>)
+; CHECK-NEXT:    ret <64 x i8> [[TMP1]]
 ;
   %1 = call <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8> <i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16, i8 undef, i8 -2, i8 -3, i8 -4, i8 -5, i8 undef, i8 -7, i8 -8, i8 -9, i8 -10, i8 -11, i8 -12, i8 -13, i8 -14, i8 -15, i8 -16>, <64 x i8> <i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, i8 undef, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17>, <64 x i8> zeroinitializer, i64 -1)
   ret <64 x i8> %1
 }
 
 define <8 x i16> @sse2_subs_w_constant() {
 ; CHECK-LABEL: @sse2_subs_w_constant(
 ; CHECK-NEXT:    ret <8 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16>

 ; CHECK-NEXT:    ret <8 x i16> <i16 -2, i16 32767, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16>
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> <i16 -1, i16 8248, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -8>, <8 x i16> <i16 1, i16 -25192, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
   ret <8 x i16> %1
 }
 
 define <8 x i16> @sse2_subs_w_constant_undefs() {
 ; CHECK-LABEL: @sse2_subs_w_constant_undefs(
-; CHECK-NEXT:    ret <8 x i16> <i16 undef, i16 -4, i16 -6, i16 -8, i16 -10, i16 undef, i16 -14, i16 -16>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> <i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8>, <8 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
+; CHECK-NEXT:    ret <8 x i16> [[TMP1]]
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> <i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8>, <8 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>)
   ret <8 x i16> %1
 }
 
 define <16 x i16> @avx2_subs_w_constant() {
 ; CHECK-LABEL: @avx2_subs_w_constant(
 ; CHECK-NEXT:    ret <16 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33>

 ; CHECK-NEXT:    ret <16 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16, i16 -18, i16 -20, i16 32767, i16 -24, i16 -26, i16 -28, i16 -30, i16 32767>
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> <i16 -1, i16 -2, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -8, i16 -9, i16 -10, i16 20125, i16 -12, i16 -13, i16 -14, i16 -15, i16 20160>, <16 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 -20230, i16 12, i16 13, i16 14, i16 15, i16 -20120>)
   ret <16 x i16> %1
 }
 
 define <16 x i16> @avx2_subs_w_constant_undefs() {
 ; CHECK-LABEL: @avx2_subs_w_constant_undefs(
-; CHECK-NEXT:    ret <16 x i16> <i16 undef, i16 -4, i16 -6, i16 -8, i16 -10, i16 undef, i16 -14, i16 -16, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> <i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <16 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
+; CHECK-NEXT:    ret <16 x i16> [[TMP1]]
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16> <i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <16 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
   ret <16 x i16> %1
 }
 
 define <32 x i16> @avx512_mask_subs_w_constant() {
 ; CHECK-LABEL: @avx512_mask_subs_w_constant(
 ; CHECK-NEXT:    ret <32 x i16> <i16 -2, i16 0, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33, i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33>

 ; CHECK-NEXT:    ret <32 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16, i16 -18, i16 -20, i16 32767, i16 -24, i16 -26, i16 -28, i16 -30, i16 32767, i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -16, i16 -18, i16 -20, i16 32767, i16 -24, i16 -26, i16 -28, i16 -30, i16 32767>
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> <i16 -1, i16 -2, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -8, i16 -9, i16 -10, i16 20125, i16 -12, i16 -13, i16 -14, i16 -15, i16 20200, i16 -1, i16 -2, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -8, i16 -9, i16 -10, i16 20125, i16 -12, i16 -13, i16 -14, i16 -15, i16 20200>, <32 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 -20200, i16 12, i16 13, i16 14, i16 15, i16 -20120, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 -20211, i16 12, i16 13, i16 14, i16 15, i16 -20120>, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %1
 }
 
 define <32 x i16> @avx512_mask_subs_w_constant_undefs() {
 ; CHECK-LABEL: @avx512_mask_subs_w_constant_undefs(
-; CHECK-NEXT:    ret <32 x i16> <i16 undef, i16 -4, i16 -6, i16 -8, i16 -10, i16 undef, i16 -14, i16 -16, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33, i16 undef, i16 -4, i16 -6, i16 -8, i16 -10, i16 undef, i16 -14, i16 -16, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33>
+; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> <i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16, i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <32 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
+; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16> <i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16, i16 undef, i16 -2, i16 -3, i16 -4, i16 -5, i16 undef, i16 -7, i16 -8, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <32 x i16> <i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 undef, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>, <32 x i16> zeroinitializer, i32 -1)
   ret <32 x i16> %1
 }
 
 declare <16 x i8> @llvm.x86.sse2.padds.b(<16 x i8>, <16 x i8>) nounwind readnone
 declare <16 x i8> @llvm.x86.sse2.psubs.b(<16 x i8>, <16 x i8>) nounwind readnone
 declare <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16>, <8 x i16>) nounwind readnone