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

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

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

Differential Revision: https://reviews.llvm.org/D55894

llvm-svn: 349892

3 files changed, 23 insertions, 125 deletions

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 7802b2dcd9b..51fef27999b 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -77,10 +77,18 @@ static bool ShouldUpgradeX86Intrinsic(Function *F, StringRef Name) {
       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
@@ -284,8 +292,6 @@ static bool ShouldUpgradeX86Intrinsic(Function *F, StringRef Name) {
       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.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
@@ -928,12 +934,14 @@ static Value *UpgradeX86ALIGNIntrinsics(IRBuilder<> &Builder, Value *Op0,
 }
 
 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});
 
@@ -1380,36 +1388,6 @@ static bool upgradeAVX512MaskToSelect(StringRef Name, IRBuilder<> &Builder,
       IID = Intrinsic::x86_avx512_vpshrd_w_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;
 
@@ -2093,6 +2071,16 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
       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.") ||
@@ -2100,7 +2088,7 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
                          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),
diff --git a/llvm/lib/Target/X86/X86IntrinsicsInfo.h b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
index dc8b3d62330..ab3fea62982 100644
--- a/llvm/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/llvm/lib/Target/X86/X86IntrinsicsInfo.h
@@ -319,8 +319,6 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   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),
@@ -361,8 +359,6 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   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),
@@ -920,8 +916,6 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   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),
@@ -980,8 +974,6 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   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),
@@ -1144,8 +1136,6 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   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),
@@ -1167,8 +1157,6 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   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),
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index 3e6a4965336..dbbe32b5636 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -248,67 +248,6 @@ Instruction *InstCombiner::SimplifyAnyMemSet(AnyMemSetInst *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;
@@ -2789,23 +2728,6 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
     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 ).