Index: llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp
===================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -576,6 +576,105 @@
   return ConstantVector::get(Vals);
 }
 
+// Replace X86-specific intrinsics with generic floor-ceil where applicable.
+static Value *simplifyX86round(IntrinsicInst &II,
+                               InstCombiner::BuilderTy &Builder) {
+  ConstantInt *Arg = nullptr;
+  Intrinsic::ID IntrinsicID = II.getIntrinsicID();
+
+  if (IntrinsicID == Intrinsic::x86_sse41_round_ss ||
+      IntrinsicID == Intrinsic::x86_sse41_round_sd)
+    Arg = dyn_cast<ConstantInt>(II.getArgOperand(2));
+  else if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ss ||
+           IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_sd)
+    Arg = dyn_cast<ConstantInt>(II.getArgOperand(4));
+  else
+    Arg = dyn_cast<ConstantInt>(II.getArgOperand(1));
+  if (!Arg)
+    return nullptr;
+  unsigned RoundControl = Arg->getZExtValue();
+
+  Arg = nullptr;
+  unsigned SAE = 0;
+  if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ps_512 ||
+      IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_pd_512)
+    Arg = dyn_cast<ConstantInt>(II.getArgOperand(4));
+  else if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ss ||
+           IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_sd)
+    Arg = dyn_cast<ConstantInt>(II.getArgOperand(5));
+  else
+    SAE = 4;
+  if (!SAE) {
+    if (!Arg)
+      return nullptr;
+    SAE = Arg->getZExtValue();
+  }
+
+  if (SAE != 4 || (RoundControl != 2 /*ceil*/ && RoundControl != 1 /*floor*/))
+    return nullptr;
+
+  Value *Src, *Dst, *Mask;
+  bool IsScalar = false;
+  if (IntrinsicID == Intrinsic::x86_sse41_round_ss ||
+      IntrinsicID == Intrinsic::x86_sse41_round_sd ||
+      IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ss ||
+      IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_sd) {
+    IsScalar = true;
+    if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ss ||
+        IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_sd) {
+      Mask = II.getArgOperand(3);
+      Value *Zero = Constant::getNullValue(Mask->getType());
+      Mask = Builder.CreateAnd(Mask, 1);
+      Mask = Builder.CreateICmp(ICmpInst::ICMP_NE, Mask, Zero);
+      Dst = II.getArgOperand(2);
+    } else
+      Dst = II.getArgOperand(0);
+    Src = Builder.CreateExtractElement(II.getArgOperand(1), (uint64_t)0);
+  } else {
+    Src = II.getArgOperand(0);
+    if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ps_128 ||
+        IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ps_256 ||
+        IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ps_512 ||
+        IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_pd_128 ||
+        IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_pd_256 ||
+        IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_pd_512) {
+      Dst = II.getArgOperand(2);
+      Mask = II.getArgOperand(3);
+    } else {
+      Dst = Src;
+      Mask = ConstantInt::getAllOnesValue(
+          Builder.getIntNTy(Src->getType()->getVectorNumElements()));
+    }
+  }
+
+  Intrinsic::ID ID = (RoundControl == 2) ? Intrinsic::ceil : Intrinsic::floor;
+  Value *Res = Builder.CreateIntrinsic(ID, {Src}, &II);
+  if (!IsScalar) {
+    if (auto *C = dyn_cast<Constant>(Mask))
+      if (C->isAllOnesValue())
+        return Res;
+    auto *MaskTy = VectorType::get(
+        Builder.getInt1Ty(), cast<IntegerType>(Mask->getType())->getBitWidth());
+    Mask = Builder.CreateBitCast(Mask, MaskTy);
+    unsigned Width = Src->getType()->getVectorNumElements();
+    if (MaskTy->getVectorNumElements() > Width) {
+      uint32_t Indices[4];
+      for (unsigned i = 0; i != Width; ++i)
+        Indices[i] = i;
+      Mask = Builder.CreateShuffleVector(Mask, Mask,
+                                         makeArrayRef(Indices, Width));
+    }
+    return Builder.CreateSelect(Mask, Res, Dst);
+  }
+  if (IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_ss ||
+      IntrinsicID == Intrinsic::x86_avx512_mask_rndscale_sd) {
+    Dst = Builder.CreateExtractElement(Dst, (uint64_t)0);
+    Res = Builder.CreateSelect(Mask, Res, Dst);
+    Dst = II.getArgOperand(0);
+  }
+  return Builder.CreateInsertElement(Dst, Res, (uint64_t)0);
+}
+
 static Value *simplifyX86movmsk(const IntrinsicInst &II) {
   Value *Arg = II.getArgOperand(0);
   Type *ResTy = II.getType();
@@ -2222,6 +2321,22 @@
     break;
   }
 
+  case Intrinsic::x86_sse41_round_ps:
+  case Intrinsic::x86_sse41_round_pd:
+  case Intrinsic::x86_avx_round_ps_256:
+  case Intrinsic::x86_avx_round_pd_256:
+  case Intrinsic::x86_avx512_mask_rndscale_ps_128:
+  case Intrinsic::x86_avx512_mask_rndscale_ps_256:
+  case Intrinsic::x86_avx512_mask_rndscale_ps_512:
+  case Intrinsic::x86_avx512_mask_rndscale_pd_128:
+  case Intrinsic::x86_avx512_mask_rndscale_pd_256:
+  case Intrinsic::x86_avx512_mask_rndscale_pd_512:
+  case Intrinsic::x86_avx512_mask_rndscale_ss:
+  case Intrinsic::x86_avx512_mask_rndscale_sd:
+    if (Value *V = simplifyX86round(*II, Builder))
+      return replaceInstUsesWith(*II, V);
+    break;
+
   case Intrinsic::x86_mmx_pmovmskb:
   case Intrinsic::x86_sse_movmsk_ps:
   case Intrinsic::x86_sse2_movmsk_pd:
@@ -2438,8 +2553,6 @@
   case Intrinsic::x86_sse2_cmp_sd:
   case Intrinsic::x86_sse2_min_sd:
   case Intrinsic::x86_sse2_max_sd:
-  case Intrinsic::x86_sse41_round_ss:
-  case Intrinsic::x86_sse41_round_sd:
   case Intrinsic::x86_xop_vfrcz_ss:
   case Intrinsic::x86_xop_vfrcz_sd: {
    unsigned VWidth = II->getType()->getVectorNumElements();
@@ -2452,6 +2565,19 @@
    }
    break;
   }
+  case Intrinsic::x86_sse41_round_ss:
+  case Intrinsic::x86_sse41_round_sd: {
+    unsigned VWidth = II->getType()->getVectorNumElements();
+    APInt UndefElts(VWidth, 0);
+    APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
+    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 ashr( <A x Bi>, Ci ).
   // Constant fold lshr( <A x Bi>, Ci ).
Index: llvm/trunk/test/Transforms/InstCombine/X86/x86-avx.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/X86/x86-avx.ll
+++ llvm/trunk/test/Transforms/InstCombine/X86/x86-avx.ll
@@ -0,0 +1,41 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+declare <8 x float> @llvm.x86.avx.round.ps.256(<8 x float>, i32)
+declare <4 x double> @llvm.x86.avx.round.pd.256(<4 x double>, i32)
+
+define <8 x float> @test_round_ps_floor(<8 x float> %a) {
+; CHECK-LABEL: @test_round_ps_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[A:%.*]])
+; CHECK-NEXT:    ret <8 x float> [[TMP1]]
+;
+  %1 = call <8 x float> @llvm.x86.avx.round.ps.256(<8 x float> %a, i32 1)
+  ret <8 x float> %1
+}
+
+define <8 x float> @test_round_ps_ceil(<8 x float> %a) {
+; CHECK-LABEL: @test_round_ps_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[A:%.*]])
+; CHECK-NEXT:    ret <8 x float> [[TMP1]]
+;
+  %1 = call <8 x float> @llvm.x86.avx.round.ps.256(<8 x float> %a, i32 2)
+  ret <8 x float> %1
+}
+
+define <4 x double> @test_round_pd_floor(<4 x double> %a) {
+; CHECK-LABEL: @test_round_pd_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[A:%.*]])
+; CHECK-NEXT:    ret <4 x double> [[TMP1]]
+;
+  %1 = call <4 x double> @llvm.x86.avx.round.pd.256(<4 x double> %a, i32 1)
+  ret <4 x double> %1
+}
+
+define <4 x double> @test_round_pd_ceil(<4 x double> %a) {
+; CHECK-LABEL: @test_round_pd_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[A:%.*]])
+; CHECK-NEXT:    ret <4 x double> [[TMP1]]
+;
+  %1 = call <4 x double> @llvm.x86.avx.round.pd.256(<4 x double> %a, i32 2)
+  ret <4 x double> %1
+}
Index: llvm/trunk/test/Transforms/InstCombine/X86/x86-avx512.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/X86/x86-avx512.ll
+++ llvm/trunk/test/Transforms/InstCombine/X86/x86-avx512.ll
@@ -916,6 +916,213 @@
 declare i32 @llvm.x86.avx512.cvttsd2usi(<2 x double>, i32)
 declare i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double>, i32)
 
+declare <4 x float> @llvm.x86.avx512.mask.rndscale.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32, i32)
+declare <2 x double> @llvm.x86.avx512.mask.rndscale.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32, i32)
+declare <4 x float> @llvm.x86.avx512.mask.rndscale.ps.128(<4 x float>, i32, <4 x float>, i8)
+declare <8 x float> @llvm.x86.avx512.mask.rndscale.ps.256(<8 x float>, i32, <8 x float>, i8)
+declare <16 x float> @llvm.x86.avx512.mask.rndscale.ps.512(<16 x float>, i32, <16 x float>, i16, i32)
+declare <2 x double> @llvm.x86.avx512.mask.rndscale.pd.128(<2 x double>, i32, <2 x double>, i8)
+declare <4 x double> @llvm.x86.avx512.mask.rndscale.pd.256(<4 x double>, i32, <4 x double>, i8)
+declare <8 x double> @llvm.x86.avx512.mask.rndscale.pd.512(<8 x double>, i32, <8 x double>, i8, i32)
+
+define <4 x float> @test_rndscale_ss_floor(<4 x float> %src0, <4 x float> %src1, <4 x float> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_ss_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i8 [[K:%.*]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[SRC1:%.*]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.floor.f32(float [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[DST:%.*]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP2]], float [[TMP5]], float [[TMP4]]
+; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[SRC0:%.*]], float [[TMP6]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP7]]
+;
+  %1 = call <4 x float> @llvm.x86.avx512.mask.rndscale.ss(<4 x float> %src0, <4 x float> %src1, <4 x float> %dst, i8 %k, i32 1, i32 4)
+  ret <4 x float> %1
+}
+
+define <4 x float> @test_rndscale_ss_ceil(<4 x float> %src0, <4 x float> %src1, <4 x float> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_ss_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i8 [[K:%.*]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[SRC1:%.*]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = call float @llvm.ceil.f32(float [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x float> [[DST:%.*]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP2]], float [[TMP5]], float [[TMP4]]
+; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <4 x float> [[SRC0:%.*]], float [[TMP6]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP7]]
+;
+  %1 = call <4 x float> @llvm.x86.avx512.mask.rndscale.ss(<4 x float> %src0, <4 x float> %src1, <4 x float> %dst, i8 %k, i32 2, i32 4)
+  ret <4 x float> %1
+}
+
+define <2 x double> @test_rndscale_sd_floor(<2 x double> %src0, <2 x double> %src1, <2 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_sd_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i8 [[K:%.*]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[SRC1:%.*]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.floor.f64(double [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[DST:%.*]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP2]], double [[TMP5]], double [[TMP4]]
+; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[SRC0:%.*]], double [[TMP6]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP7]]
+;
+  %1 = call <2 x double> @llvm.x86.avx512.mask.rndscale.sd(<2 x double> %src0, <2 x double> %src1, <2 x double> %dst, i8 %k, i32 1, i32 4)
+  ret <2 x double> %1
+}
+
+define <2 x double> @test_rndscale_sd_ceil(<2 x double> %src0, <2 x double> %src1, <2 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_sd_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i8 [[K:%.*]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i8 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[SRC1:%.*]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = call double @llvm.ceil.f64(double [[TMP3]])
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[DST:%.*]], i64 0
+; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[TMP2]], double [[TMP5]], double [[TMP4]]
+; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[SRC0:%.*]], double [[TMP6]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP7]]
+;
+  %1 = call <2 x double> @llvm.x86.avx512.mask.rndscale.sd(<2 x double> %src0, <2 x double> %src1, <2 x double> %dst, i8 %k, i32 2, i32 4)
+  ret <2 x double> %1
+}
+
+define <4 x float> @test_rndscale_ps_128_floor(<4 x float> %src, <4 x float> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_ps_128_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x float> @llvm.floor.v4f32(<4 x float> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x float> [[TMP1]], <4 x float> [[DST:%.*]]
+; CHECK-NEXT:    ret <4 x float> [[TMP4]]
+;
+  %1 = call <4 x float> @llvm.x86.avx512.mask.rndscale.ps.128(<4 x float> %src, i32 1, <4 x float> %dst, i8 %k)
+  ret <4 x float> %1
+}
+
+define <4 x float> @test_rndscale_ps_128_ceil(<4 x float> %src, <4 x float> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_ps_128_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x float> @llvm.ceil.v4f32(<4 x float> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x float> [[TMP1]], <4 x float> [[DST:%.*]]
+; CHECK-NEXT:    ret <4 x float> [[TMP4]]
+;
+  %1 = call <4 x float> @llvm.x86.avx512.mask.rndscale.ps.128(<4 x float> %src, i32 2, <4 x float> %dst, i8 %k)
+  ret <4 x float> %1
+}
+
+define <8 x float> @test_rndscale_ps_256_floor(<8 x float> %src, <8 x float> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_ps_256_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x float> @llvm.floor.v8f32(<8 x float> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x float> [[TMP1]], <8 x float> [[DST:%.*]]
+; CHECK-NEXT:    ret <8 x float> [[TMP3]]
+;
+  %1 = call <8 x float> @llvm.x86.avx512.mask.rndscale.ps.256(<8 x float> %src, i32 1, <8 x float> %dst, i8 %k)
+  ret <8 x float> %1
+}
+
+define <8 x float> @test_rndscale_ps_256_ceil(<8 x float> %src, <8 x float> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_ps_256_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x float> @llvm.ceil.v8f32(<8 x float> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x float> [[TMP1]], <8 x float> [[DST:%.*]]
+; CHECK-NEXT:    ret <8 x float> [[TMP3]]
+;
+  %1 = call <8 x float> @llvm.x86.avx512.mask.rndscale.ps.256(<8 x float> %src, i32 2, <8 x float> %dst, i8 %k)
+  ret <8 x float> %1
+}
+
+define <16 x float> @test_rndscale_ps_512_floor(<16 x float> %src, <16 x float> %dst, i16 %k) {
+; CHECK-LABEL: @test_rndscale_ps_512_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x float> @llvm.floor.v16f32(<16 x float> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i16 [[K:%.*]] to <16 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <16 x i1> [[TMP2]], <16 x float> [[TMP1]], <16 x float> [[DST:%.*]]
+; CHECK-NEXT:    ret <16 x float> [[TMP3]]
+;
+  %1 = call <16 x float> @llvm.x86.avx512.mask.rndscale.ps.512(<16 x float> %src, i32 1, <16 x float> %dst, i16 %k, i32 4)
+  ret <16 x float> %1
+}
+
+define <16 x float> @test_rndscale_ps_512_ceil(<16 x float> %src, <16 x float> %dst, i16 %k) {
+; CHECK-LABEL: @test_rndscale_ps_512_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x float> @llvm.ceil.v16f32(<16 x float> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i16 [[K:%.*]] to <16 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <16 x i1> [[TMP2]], <16 x float> [[TMP1]], <16 x float> [[DST:%.*]]
+; CHECK-NEXT:    ret <16 x float> [[TMP3]]
+;
+  %1 = call <16 x float> @llvm.x86.avx512.mask.rndscale.ps.512(<16 x float> %src, i32 2, <16 x float> %dst, i16 %k, i32 4)
+  ret <16 x float> %1
+}
+
+define <2 x double> @test_rndscale_pd_128_floor(<2 x double> %src, <2 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_pd_128_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <2 x double> @llvm.floor.v2f64(<2 x double> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP4:%.*]] = select <2 x i1> [[TMP3]], <2 x double> [[TMP1]], <2 x double> [[DST:%.*]]
+; CHECK-NEXT:    ret <2 x double> [[TMP4]]
+;
+  %1 = call <2 x double> @llvm.x86.avx512.mask.rndscale.pd.128(<2 x double> %src, i32 1, <2 x double> %dst, i8 %k)
+  ret <2 x double> %1
+}
+
+define <2 x double> @test_rndscale_pd_128_ceil(<2 x double> %src, <2 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_pd_128_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <2 x double> @llvm.ceil.v2f64(<2 x double> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP4:%.*]] = select <2 x i1> [[TMP3]], <2 x double> [[TMP1]], <2 x double> [[DST:%.*]]
+; CHECK-NEXT:    ret <2 x double> [[TMP4]]
+;
+  %1 = call <2 x double> @llvm.x86.avx512.mask.rndscale.pd.128(<2 x double> %src, i32 2, <2 x double> %dst, i8 %k)
+  ret <2 x double> %1
+}
+
+define <4 x double> @test_rndscale_pd_256_floor(<4 x double> %src, <4 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_pd_256_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x double> @llvm.floor.v4f64(<4 x double> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x double> [[TMP1]], <4 x double> [[DST:%.*]]
+; CHECK-NEXT:    ret <4 x double> [[TMP4]]
+;
+  %1 = call <4 x double> @llvm.x86.avx512.mask.rndscale.pd.256(<4 x double> %src, i32 1, <4 x double> %dst, i8 %k)
+  ret <4 x double> %1
+}
+
+define <4 x double> @test_rndscale_pd_256_ceil(<4 x double> %src, <4 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_pd_256_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x double> @llvm.ceil.v4f64(<4 x double> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x double> [[TMP1]], <4 x double> [[DST:%.*]]
+; CHECK-NEXT:    ret <4 x double> [[TMP4]]
+;
+  %1 = call <4 x double> @llvm.x86.avx512.mask.rndscale.pd.256(<4 x double> %src, i32 2, <4 x double> %dst, i8 %k)
+  ret <4 x double> %1
+}
+
+define <8 x double> @test_rndscale_pd_512_floor(<8 x double> %src, <8 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_pd_512_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x double> @llvm.floor.v8f64(<8 x double> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x double> [[TMP1]], <8 x double> [[DST:%.*]]
+; CHECK-NEXT:    ret <8 x double> [[TMP3]]
+;
+  %1 = call <8 x double> @llvm.x86.avx512.mask.rndscale.pd.512(<8 x double> %src, i32 1, <8 x double> %dst, i8 %k, i32 4)
+  ret <8 x double> %1
+}
+
+define <8 x double> @test_rndscale_pd_512_ceil(<8 x double> %src, <8 x double> %dst, i8 %k) {
+; CHECK-LABEL: @test_rndscale_pd_512_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x double> @llvm.ceil.v8f64(<8 x double> [[SRC:%.*]])
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8 [[K:%.*]] to <8 x i1>
+; CHECK-NEXT:    [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x double> [[TMP1]], <8 x double> [[DST:%.*]]
+; CHECK-NEXT:    ret <8 x double> [[TMP3]]
+;
+  %1 = call <8 x double> @llvm.x86.avx512.mask.rndscale.pd.512(<8 x double> %src, i32 2, <8 x double> %dst, i8 %k, i32 4)
+  ret <8 x double> %1
+}
+
 declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)
 
 define <4 x float> @test_mask_vfmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
Index: llvm/trunk/test/Transforms/InstCombine/X86/x86-sse41.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/X86/x86-sse41.ll
+++ llvm/trunk/test/Transforms/InstCombine/X86/x86-sse41.ll
@@ -13,6 +13,28 @@
   ret <2 x double> %3
 }
 
+define <2 x double> @test_round_sd_floor(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: @test_round_sd_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = call double @llvm.floor.f64(double [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> [[A:%.*]], double [[TMP2]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP3]]
+;
+  %1 = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> %a, <2 x double> %b, i32 1)
+  ret <2 x double> %1
+}
+
+define <2 x double> @test_round_sd_ceil(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: @test_round_sd_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> [[B:%.*]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = call double @llvm.ceil.f64(double [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> [[A:%.*]], double [[TMP2]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP3]]
+;
+  %1 = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> %a, <2 x double> %b, i32 2)
+  ret <2 x double> %1
+}
+
 define double @test_round_sd_0(double %a, double %b) {
 ; CHECK-LABEL: @test_round_sd_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double [[B:%.*]], i32 0
@@ -57,6 +79,28 @@
   ret <4 x float> %7
 }
 
+define <4 x float> @test_round_ss_floor(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @test_round_ss_floor(
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = call float @llvm.floor.f32(float [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[A:%.*]], float [[TMP2]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP3]]
+;
+  %1 = tail call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> %a, <4 x float> %b, i32 1)
+  ret <4 x float> %1
+}
+
+define <4 x float> @test_round_ss_ceil(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @test_round_ss_ceil(
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[B:%.*]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = call float @llvm.ceil.f32(float [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[A:%.*]], float [[TMP2]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP3]]
+;
+  %1 = tail call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> %a, <4 x float> %b, i32 2)
+  ret <4 x float> %1
+}
+
 define float @test_round_ss_0(float %a, float %b) {
 ; CHECK-LABEL: @test_round_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float [[B:%.*]], i32 0