[X86] Constant folding of adds/subs intrinsics

[tkrupa]

Summary

This was originally a part of D46179 but got so different to unsigned versions that it's better to have it in a separate patch.

[spatel]

The title of this patch includes "Constant folding" - shouldn't the code be in ConstantFolding.cpp or some descendant of that?

I remember that there was a suggestion (years ago?) to split target-specific logic like this into its own files because other targets shouldn't be burdened with code that will never affect them, but this hasn't happened yet. There's x86-specific intrinsic code in ConstantFolding.cpp added with rL123206.

[tkrupa]

I added it to InstCombineCalls.cpp after @craig.topper's suggestion to do so in order to enable adding more optimizations besides constant folding in the same place.

[spatel]

I'm not finding the mails where the extraction of target-specific code came up, but my vague memory says @rnk had some ideas about how to improve things.

[craig.topper]

lib/Transforms/InstCombine/InstCombineCalls.cpp

265

We should really remove masking from these intrinsics and use a select. But that can be a follow up.

286

Probably could use getIntegerBitWidth instead of getPrimitiveSizeInBits. Should be slightly cheaper since getPrimitiveSizeInBits has to detect that its an integer and then call getIntegerBitWidth.

296

const APInt & Val0

296

Isn't it possible for the element to be a ConstantExpr? In which case this case would fail

[tkrupa]

Implemented suggested changes.
Regarding masking with select - do you mean creating new avx512_padds/avx512_psubs intrinsics without masks and replacing the old calls with new intrinsic+select?

[craig.topper]

LGTM with those two formatting fixes.

lib/Transforms/InstCombine/InstCombineCalls.cpp

288

There's an extra space after 'unsigned'

305

This is overindented.

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 = false;
+  bool IsMasked = false;
+
+  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:
+    IsAddition = true; IsMasked = false;
+    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:
+    IsAddition = false; IsMasked = false;
+    break;
+  case Intrinsic::x86_avx512_mask_padds_b_512:
+  case Intrinsic::x86_avx512_mask_padds_w_512:
+    IsAddition = true; IsMasked = true;
+    break;
+  case Intrinsic::x86_avx512_mask_psubs_b_512:
+  case Intrinsic::x86_avx512_mask_psubs_w_512:
+    IsAddition = false; IsMasked = true;
+    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 || (IsMasked && !isa<Constant>(II.getOperand(2))))
+    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));
+  }
+
+  Value *ResultVec = ConstantVector::get(Result);
+
+  if (II.getNumArgOperands() == 4) { // For masked intrinsics.
+    Value *Src = II.getOperand(2);
+    auto Mask = II.getOperand(3);
+    if (auto *C = dyn_cast<Constant>(Mask))
+      if (C->isAllOnesValue())
+        return ResultVec;
+    auto *MaskTy = VectorType::get(
+        Builder.getInt1Ty(), cast<IntegerType>(Mask->getType())->getBitWidth());
+    Mask = Builder.CreateBitCast(Mask, MaskTy);
+    ResultVec = Builder.CreateSelect(Mask, ResultVec, Src);
+  }
+
+  return ResultVec;
+}
+
 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_mask_padds_b_512:
+  case Intrinsic::x86_avx512_mask_padds_w_512:
+  case Intrinsic::x86_avx512_mask_psubs_b_512:
+  case Intrinsic::x86_avx512_mask_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/Transforms/InstCombine/X86/x86-adds-subs.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+define <16 x i8> @sse2_adds_b_constant() {
+; CHECK-LABEL: @sse2_adds_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>
+  %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 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 8, 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_underflow() {
+; CHECK-LABEL: @sse2_adds_b_constant_underflow(
+; 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>
+  %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>
+  %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 11, 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 11, 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 17, 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 17>)
+  ret <32 x i8> %1
+}
+
+define <32 x i8> @avx2_adds_b_constant_underflow() {
+; CHECK-LABEL: @avx2_adds_b_constant_underflow(
+; CHECK-NEXT: ret <32 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, 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 <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 -107, i8 9, i8 10, i8 11, 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 -107, i8 9, i8 10, i8 11, 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 -48, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, 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 <32 x i8> %1
+}
+
+define <32 x i8> @avx2_adds_b_constant_overflow() {
+; CHECK-LABEL: @avx2_adds_b_constant_overflow(
+; 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>
+  %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>
+  %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 11, 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 11, 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 11, 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 11, 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 17, 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 17, 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 17, 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 17>, <64 x i8> zeroinitializer, i64 -3)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @avx512_mask_adds_b_constant_underflow() {
+; CHECK-LABEL: @avx512_mask_adds_b_constant_underflow(
+; CHECK-NEXT: ret <64 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, 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, 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, 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 <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 -107, i8 9, i8 10, i8 11, 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 -107, i8 9, i8 10, i8 11, 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 -107, i8 9, i8 10, i8 11, 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 -107, i8 9, i8 10, i8 11, 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 -48, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, 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, 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, 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>, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @avx512_mask_adds_b_constant_overflow() {
+; CHECK-LABEL: @avx512_mask_adds_b_constant_overflow(
+; 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>
+  %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>
+  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8>, <8 x i16> <i16 1, i16 2, 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_underflow() {
+; CHECK-LABEL: @sse2_adds_w_constant_underflow(
+; CHECK-NEXT: ret <8 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 -32768>
+  %1 = call <8 x i16> @llvm.x86.sse2.padds.w(<8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 -32107>, <8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 -12188>)
+  ret <8 x i16> %1
+}
+
+define <8 x i16> @sse2_adds_w_constant_overflow() {
+; CHECK-LABEL: @sse2_adds_w_constant_overflow(
+; 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>
+  %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>
+  %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 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <16 x i16> <i16 1, 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 <16 x i16> @avx2_adds_w_constant_underflow() {
+; CHECK-LABEL: @avx2_adds_w_constant_underflow(
+; CHECK-NEXT: ret <16 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 -32768, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33>
+  %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 -21107, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <16 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 -15188, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
+  ret <16 x i16> %1
+}
+
+define <16 x i16> @avx2_adds_w_constant_overflow() {
+; CHECK-LABEL: @avx2_adds_w_constant_overflow(
+; 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>
+  %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>
+  %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 11, i16 12, i16 13, i16 14, i16 15, i16 16, i16 1, 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 16>, <32 x i16> <i16 1, 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 1, 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 -3)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @avx512_mask_adds_w_constant_underflow() {
+; CHECK-LABEL: @avx512_mask_adds_w_constant_underflow(
+; CHECK-NEXT: ret <32 x i16> <i16 2, i16 4, i16 6, i16 8, i16 10, i16 12, i16 14, i16 -32768, 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 -32768, i16 18, i16 20, i16 22, i16 24, i16 26, i16 28, i16 30, i16 33>
+  %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 -20107, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 -20107, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 16>, <32 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 -20168, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 -20248, 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 <32 x i16> @avx512_mask_adds_w_constant_overflow() {
+; CHECK-LABEL: @avx512_mask_adds_w_constant_overflow(
+; 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>
+  %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>
+  %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 -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 8, 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_subs_b_constant_underflow() {
+; CHECK-LABEL: @sse2_subs_b_constant_underflow(
+; 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.psubs.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_subs_b_constant_overflow() {
+; CHECK-LABEL: @sse2_subs_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.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>
+  %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>
+  %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 -11, 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 -11, 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 17, 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 17>)
+  ret <32 x i8> %1
+}
+
+define <32 x i8> @avx2_subs_b_constant_underflow() {
+; CHECK-LABEL: @avx2_subs_b_constant_underflow(
+; CHECK-NEXT: ret <32 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, 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 <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 -107, i8 -9, i8 -10, i8 -11, 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 -107, i8 -9, i8 -10, i8 -11, 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 48, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, 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 <32 x i8> %1
+}
+
+define <32 x i8> @avx2_subs_b_constant_overflow() {
+; CHECK-LABEL: @avx2_subs_b_constant_overflow(
+; 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>
+  %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>
+  %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 -11, 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 -11, 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 -11, 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 -11, 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 17, 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 17, 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 17, 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 17>, <64 x i8> zeroinitializer, i64 -3)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @avx512_mask_subs_b_constant_underflow() {
+; CHECK-LABEL: @avx512_mask_subs_b_constant_underflow(
+; CHECK-NEXT: ret <64 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, 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, 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, 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 <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 -107, i8 -9, i8 -10, i8 -11, 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 -107, i8 -9, i8 -10, i8 -11, 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 -107, i8 -9, i8 -10, i8 -11, 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 -107, i8 -9, i8 -10, i8 -11, 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 48, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 17, 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, 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, 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>, <64 x i8> zeroinitializer, i64 -1)
+  ret <64 x i8> %1
+}
+
+define <64 x i8> @avx512_mask_subs_b_constant_overflow() {
+; CHECK-LABEL: @avx512_mask_subs_b_constant_overflow(
+; 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>
+  %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>
+  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> <i16 -1, i16 -2, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -8>, <8 x i16> <i16 1, i16 2, 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_underflow() {
+; CHECK-LABEL: @sse2_subs_w_constant_underflow(
+; CHECK-NEXT: ret <8 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -32768>
+  %1 = call <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16> <i16 -1, i16 -2, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -32107>, <8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 12188>)
+  ret <8 x i16> %1
+}
+
+define <8 x i16> @sse2_subs_w_constant_overflow() {
+; CHECK-LABEL: @sse2_subs_w_constant_overflow(
+; 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>
+  %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>
+  %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 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <16 x i16> <i16 1, 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 <16 x i16> @avx2_subs_w_constant_underflow() {
+; CHECK-LABEL: @avx2_subs_w_constant_underflow(
+; CHECK-NEXT: ret <16 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -32768, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33>
+  %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 -21107, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <16 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 15188, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17>)
+  ret <16 x i16> %1
+}
+
+define <16 x i16> @avx2_subs_w_constant_overflow() {
+; CHECK-LABEL: @avx2_subs_w_constant_overflow(
+; 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>
+  %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>
+  %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 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16, i16 -1, 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 -16>, <32 x i16> <i16 1, 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 1, 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 -3)
+  ret <32 x i16> %1
+}
+
+define <32 x i16> @avx512_mask_subs_w_constant_underflow() {
+; CHECK-LABEL: @avx512_mask_subs_w_constant_underflow(
+; CHECK-NEXT: ret <32 x i16> <i16 -2, i16 -4, i16 -6, i16 -8, i16 -10, i16 -12, i16 -14, i16 -32768, 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 -32768, i16 -18, i16 -20, i16 -22, i16 -24, i16 -26, i16 -28, i16 -30, i16 -33>
+  %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 -20107, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16, i16 -1, i16 -2, i16 -3, i16 -4, i16 -5, i16 -6, i16 -7, i16 -20107, i16 -9, i16 -10, i16 -11, i16 -12, i16 -13, i16 -14, i16 -15, i16 -16>, <32 x i16> <i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 20168, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 17, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 20248, 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 <32 x i16> @avx512_mask_subs_w_constant_overflow() {
+; CHECK-LABEL: @avx512_mask_subs_w_constant_overflow(
+; 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>
+  %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
+declare <8 x i16> @llvm.x86.sse2.psubs.w(<8 x i16>, <8 x i16>) nounwind readnone
+declare <32 x i8> @llvm.x86.avx2.padds.b(<32 x i8>, <32 x i8>) nounwind readnone
+declare <32 x i8> @llvm.x86.avx2.psubs.b(<32 x i8>, <32 x i8>) nounwind readnone
+declare <16 x i16> @llvm.x86.avx2.padds.w(<16 x i16>, <16 x i16>) nounwind readnone
+declare <16 x i16> @llvm.x86.avx2.psubs.w(<16 x i16>, <16 x i16>) nounwind readnone
+declare <64 x i8> @llvm.x86.avx512.mask.padds.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64) nounwind readnone
+declare <64 x i8> @llvm.x86.avx512.mask.psubs.b.512(<64 x i8>, <64 x i8>, <64 x i8>, i64) nounwind readnone
+declare <32 x i16> @llvm.x86.avx512.mask.padds.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32) nounwind readnone
+declare <32 x i16> @llvm.x86.avx512.mask.psubs.w.512(<32 x i16>, <32 x i16>, <32 x i16>, i32) nounwind readnone