[X86] Promote vXi1 fp_to_uint/fp_to_sint to vXi32 to avoid scalarization.

llvm-svn: 321632

2 files changed, 137 insertions, 3417 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 58a48985bfb..304ff21bb38 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1160,6 +1160,17 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::SINT_TO_FP,         MVT::v2i1,  Custom);
     setOperationAction(ISD::UINT_TO_FP,         MVT::v2i1,  Custom);
 
+    setOperationPromotedToType(ISD::FP_TO_SINT, MVT::v16i1, MVT::v16i32);
+    setOperationPromotedToType(ISD::FP_TO_UINT, MVT::v16i1, MVT::v16i32);
+    setOperationPromotedToType(ISD::FP_TO_SINT, MVT::v8i1,  MVT::v8i32);
+    setOperationPromotedToType(ISD::FP_TO_UINT, MVT::v8i1,  MVT::v8i32);
+    setOperationPromotedToType(ISD::FP_TO_SINT, MVT::v4i1,  MVT::v4i32);
+    setOperationPromotedToType(ISD::FP_TO_UINT, MVT::v4i1,  MVT::v4i32);
+    if (Subtarget.hasVLX()) {
+      setOperationAction(ISD::FP_TO_SINT,         MVT::v2i1,  Custom);
+      setOperationAction(ISD::FP_TO_UINT,         MVT::v2i1,  Custom);
+    }
+
     // Extends of v16i1/v8i1 to 128-bit vectors.
     setOperationAction(ISD::SIGN_EXTEND,        MVT::v16i8, Custom);
     setOperationAction(ISD::ZERO_EXTEND,        MVT::v16i8, Custom);
@@ -16671,9 +16682,29 @@ SDValue X86TargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG) const {
   MVT VT = Op.getSimpleValueType();
 
   if (VT.isVector()) {
-    assert(Subtarget.hasDQI() && Subtarget.hasVLX() && "Requires AVX512DQVL!");
     SDValue Src = Op.getOperand(0);
     SDLoc dl(Op);
+
+    if (VT == MVT::v2i1 && Src.getSimpleValueType() == MVT::v2f64) {
+      MVT ResVT = MVT::v4i32;
+      MVT TruncVT = MVT::v4i1;
+      unsigned Opc = IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI;
+      if (!IsSigned && !Subtarget.hasVLX()) {
+        // Widen to 512-bits.
+        ResVT = MVT::v8i32;
+        TruncVT = MVT::v8i1;
+        Opc = ISD::FP_TO_UINT;
+        Src = DAG.getNode(ISD::INSERT_SUBVECTOR, dl, MVT::v8f64,
+                          DAG.getUNDEF(MVT::v8f64),
+                          Src, DAG.getIntPtrConstant(0, dl));
+      }
+      SDValue Res = DAG.getNode(Opc, dl, ResVT, Src);
+      Res = DAG.getNode(ISD::TRUNCATE, dl, TruncVT, Res);
+      return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MVT::v2i1, Res,
+                         DAG.getIntPtrConstant(0, dl));
+    }
+
+    assert(Subtarget.hasDQI() && Subtarget.hasVLX() && "Requires AVX512DQVL!");
     if (VT == MVT::v2i64 && Src.getSimpleValueType() == MVT::v2f32) {
       return DAG.getNode(IsSigned ? X86ISD::CVTTP2SI : X86ISD::CVTTP2UI, dl, VT,
                          DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v4f32, Src,
diff --git a/llvm/test/CodeGen/X86/avx512-cvt.ll b/llvm/test/CodeGen/X86/avx512-cvt.ll
index c657b10fbf2..df66935b75b 100644
--- a/llvm/test/CodeGen/X86/avx512-cvt.ll
+++ b/llvm/test/CodeGen/X86/avx512-cvt.ll
@@ -2034,43 +2034,13 @@ define <2 x i64> @test_2f64toub(<2 x double> %a, <2 x i64> %passthru) {
 ; NOVLDQ-NEXT:    vzeroupper
 ; NOVLDQ-NEXT:    retq
 ;
-; VLBW-LABEL: test_2f64toub:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    andl $1, %eax
-; VLBW-NEXT:    kmovw %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLBW-NEXT:    korw %k1, %k2, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k1
-; VLBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_2f64toub:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    andl $1, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLNOBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLNOBW-NEXT:    korw %k1, %k2, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k1
-; VLNOBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
+; VL-LABEL: test_2f64toub:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttpd2udq %xmm0, %xmm0
+; VL-NEXT:    vpslld $31, %xmm0, %xmm0
+; VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
+; VL-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
+; VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_2f64toub:
 ; AVX512DQ:       # %bb.0:
@@ -2097,534 +2067,34 @@ define <4 x i64> @test_4f64toub(<4 x double> %a, <4 x i64> %passthru) {
 ; NOVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; NOVL-NEXT:    retq
 ;
-; VLBW-LABEL: test_4f64toub:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    andl $1, %eax
-; VLBW-NEXT:    kmovw %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLBW-NEXT:    korw %k1, %k2, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k1
-; VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_4f64toub:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    andl $1, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLNOBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLNOBW-NEXT:    korw %k1, %k2, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLNOBW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k1
-; VLNOBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
+; VL-LABEL: test_4f64toub:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; VL-NEXT:    vpslld $31, %xmm0, %xmm0
+; VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
+; VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptoui <4 x double> %a to <4 x i1>
   %select = select <4 x i1> %mask, <4 x i64> %passthru, <4 x i64> zeroinitializer
   ret <4 x i64> %select
 }
 
 define <8 x i64> @test_8f64toub(<8 x double> %a, <8 x i64> %passthru) {
-; KNL-LABEL: test_8f64toub:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k0
-; KNL-NEXT:    vcvttsd2si %xmm0, %eax
-; KNL-NEXT:    andl $1, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftrw $1, %k0, %k2
-; KNL-NEXT:    kshiftlw $1, %k2, %k2
-; KNL-NEXT:    korw %k1, %k2, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k2
-; KNL-NEXT:    kxorw %k0, %k2, %k0
-; KNL-NEXT:    kshiftlw $15, %k0, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k0
-; KNL-NEXT:    kxorw %k1, %k0, %k0
-; KNL-NEXT:    kshiftrw $2, %k0, %k1
-; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $13, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $3, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $12, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $4, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $11, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $5, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $10, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $6, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; KNL-NEXT:    vcvttsd2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $9, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $7, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $8, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k1
-; KNL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; KNL-NEXT:    retq
-;
-; SKX-LABEL: test_8f64toub:
-; SKX:       # %bb.0:
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k0
-; SKX-NEXT:    vcvttsd2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k2
-; SKX-NEXT:    kxorb %k0, %k2, %k0
-; SKX-NEXT:    kshiftlb $7, %k0, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k0
-; SKX-NEXT:    kxorb %k1, %k0, %k0
-; SKX-NEXT:    kshiftrb $2, %k0, %k1
-; SKX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $5, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $3, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $4, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $4, %k0, %k1
-; SKX-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $3, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $5, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $2, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k1
-; SKX-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; SKX-NEXT:    vcvttsd2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftlb $1, %k0, %k0
-; SKX-NEXT:    kshiftrb $1, %k0, %k0
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    korb %k1, %k0, %k1
-; SKX-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; SKX-NEXT:    retq
-;
-; AVX512VL-LABEL: test_8f64toub:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k0
-; AVX512VL-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VL-NEXT:    andl $1, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VL-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VL-NEXT:    korw %k1, %k2, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VL-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VL-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VL-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512VL-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VL-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_8f64toub:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k0
-; AVX512DQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512DQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512DQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512DQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512DQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    korb %k1, %k0, %k1
-; AVX512DQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512DQ-NEXT:    retq
-;
-; AVX512BW-LABEL: test_8f64toub:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k0
-; AVX512BW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512BW-NEXT:    andl $1, %eax
-; AVX512BW-NEXT:    kmovw %eax, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512BW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512BW-NEXT:    korw %k1, %k2, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512BW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512BW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512BW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    retq
-;
-; AVX512VLDQ-LABEL: test_8f64toub:
-; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k0
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512VLDQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512VLDQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    korb %k1, %k0, %k1
-; AVX512VLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLDQ-NEXT:    retq
+; NOVL-LABEL: test_8f64toub:
+; NOVL:       # %bb.0:
+; NOVL-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; NOVL-NEXT:    vpslld $31, %ymm0, %ymm0
+; NOVL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; NOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; NOVL-NEXT:    retq
 ;
-; AVX512VLBW-LABEL: test_8f64toub:
-; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k0
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLBW-NEXT:    andl $1, %eax
-; AVX512VLBW-NEXT:    kmovw %eax, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VLBW-NEXT:    korw %k1, %k2, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VLBW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VLBW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLBW-NEXT:    retq
+; VL-LABEL: test_8f64toub:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; VL-NEXT:    vpslld $31, %ymm0, %ymm0
+; VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
+; VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptoui <8 x double> %a to <8 x i1>
   %select = select <8 x i1> %mask, <8 x i64> %passthru, <8 x i64> zeroinitializer
   ret <8 x i64> %select
@@ -2698,1169 +2168,47 @@ define <4 x i64> @test_4f32toub(<4 x float> %a, <4 x i64> %passthru) {
 ; NOVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; NOVL-NEXT:    retq
 ;
-; VLBW-LABEL: test_4f32toub:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    andl $1, %eax
-; VLBW-NEXT:    kmovw %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLBW-NEXT:    korw %k1, %k2, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k1
-; VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_4f32toub:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    andl $1, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLNOBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLNOBW-NEXT:    korw %k1, %k2, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k1
-; VLNOBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
+; VL-LABEL: test_4f32toub:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttps2dq %xmm0, %xmm0
+; VL-NEXT:    vpslld $31, %xmm0, %xmm0
+; VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
+; VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptoui <4 x float> %a to <4 x i1>
   %select = select <4 x i1> %mask, <4 x i64> %passthru, <4 x i64> zeroinitializer
   ret <4 x i64> %select
 }
 
 define <8 x i64> @test_8f32toub(<8 x float> %a, <8 x i64> %passthru) {
-; KNL-LABEL: test_8f32toub:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    andl $1, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftrw $1, %k0, %k2
-; KNL-NEXT:    kshiftlw $1, %k2, %k2
-; KNL-NEXT:    korw %k1, %k2, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k2
-; KNL-NEXT:    kxorw %k0, %k2, %k0
-; KNL-NEXT:    kshiftlw $15, %k0, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k0
-; KNL-NEXT:    kxorw %k1, %k0, %k0
-; KNL-NEXT:    kshiftrw $2, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $13, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $3, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $12, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $4, %k0, %k1
-; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $11, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $5, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $10, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $6, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $9, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $7, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $8, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k1
-; KNL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; KNL-NEXT:    retq
-;
-; SKX-LABEL: test_8f32toub:
-; SKX:       # %bb.0:
-; SKX-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k0
-; SKX-NEXT:    vcvttss2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k2
-; SKX-NEXT:    kxorb %k0, %k2, %k0
-; SKX-NEXT:    kshiftlb $7, %k0, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k0
-; SKX-NEXT:    kxorb %k1, %k0, %k0
-; SKX-NEXT:    kshiftrb $2, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $5, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $3, %k0, %k1
-; SKX-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $4, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $4, %k0, %k1
-; SKX-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; SKX-NEXT:    vcvttss2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $3, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $5, %k0, %k1
-; SKX-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $2, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftlb $1, %k0, %k0
-; SKX-NEXT:    kshiftrb $1, %k0, %k0
-; SKX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; SKX-NEXT:    vcvttss2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    korb %k1, %k0, %k1
-; SKX-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; SKX-NEXT:    retq
-;
-; AVX512VL-LABEL: test_8f32toub:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k0
-; AVX512VL-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VL-NEXT:    andl $1, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VL-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VL-NEXT:    korw %k1, %k2, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VL-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VL-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VL-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512VL-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VL-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_8f32toub:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512DQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512DQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512DQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    korb %k1, %k0, %k1
-; AVX512DQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512DQ-NEXT:    retq
-;
-; AVX512BW-LABEL: test_8f32toub:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    andl $1, %eax
-; AVX512BW-NEXT:    kmovw %eax, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512BW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512BW-NEXT:    korw %k1, %k2, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512BW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512BW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512BW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    retq
-;
-; AVX512VLDQ-LABEL: test_8f32toub:
-; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k0
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512VLDQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512VLDQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512VLDQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    korb %k1, %k0, %k1
-; AVX512VLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLDQ-NEXT:    retq
+; NOVL-LABEL: test_8f32toub:
+; NOVL:       # %bb.0:
+; NOVL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; NOVL-NEXT:    vpslld $31, %ymm0, %ymm0
+; NOVL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; NOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; NOVL-NEXT:    retq
 ;
-; AVX512VLBW-LABEL: test_8f32toub:
-; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k0
-; AVX512VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLBW-NEXT:    andl $1, %eax
-; AVX512VLBW-NEXT:    kmovw %eax, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VLBW-NEXT:    korw %k1, %k2, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VLBW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VLBW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLBW-NEXT:    retq
+; VL-LABEL: test_8f32toub:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; VL-NEXT:    vpslld $31, %ymm0, %ymm0
+; VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
+; VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptoui <8 x float> %a to <8 x i1>
   %select = select <8 x i1> %mask, <8 x i64> %passthru, <8 x i64> zeroinitializer
   ret <8 x i64> %select
 }
 
 define <16 x i32> @test_16f32toub(<16 x float> %a, <16 x i32> %passthru) {
-; KNL-LABEL: test_16f32toub:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k2
-; KNL-NEXT:    kxorw %k0, %k2, %k0
-; KNL-NEXT:    kshiftlw $15, %k0, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k0
-; KNL-NEXT:    kxorw %k1, %k0, %k0
-; KNL-NEXT:    kshiftrw $2, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $13, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $3, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $12, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $4, %k0, %k1
-; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $11, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $5, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $10, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $6, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $9, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $7, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $8, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $8, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $7, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $9, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $6, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $10, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $5, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $11, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $4, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $12, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $3, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $13, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $2, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftlw $1, %k0, %k0
-; KNL-NEXT:    kshiftrw $1, %k0, %k0
-; KNL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    korw %k1, %k0, %k1
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; KNL-NEXT:    retq
-;
-; VLBW-LABEL: test_16f32toub:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $4, %k0, %k1
-; VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $11, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $5, %k0, %k1
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $10, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $6, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $9, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $7, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $8, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $8, %k0, %k1
-; VLBW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $7, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $9, %k0, %k1
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $6, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $10, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $5, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $11, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $4, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $12, %k0, %k1
-; VLBW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $3, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $13, %k0, %k1
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $2, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftlw $1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $1, %k0, %k0
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    korw %k1, %k0, %k1
-; VLBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_16f32toub:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $4, %k0, %k1
-; VLNOBW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $11, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $5, %k0, %k1
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $10, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $6, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $9, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $7, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $8, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $8, %k0, %k1
-; VLNOBW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $7, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $9, %k0, %k1
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $6, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $10, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $5, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $11, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $4, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $12, %k0, %k1
-; VLNOBW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $3, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $13, %k0, %k1
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $2, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftlw $1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k0
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    korw %k1, %k0, %k1
-; VLNOBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_16f32toub:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512DQ-NEXT:    kxorw %k0, %k2, %k0
-; AVX512DQ-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512DQ-NEXT:    kxorw %k1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $7, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $9, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $6, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $10, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $5, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $11, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $4, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $12, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $3, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $13, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $2, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $14, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $1, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftlw $1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrw $1, %k0, %k0
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    korw %k1, %k0, %k1
-; AVX512DQ-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; AVX512DQ-NEXT:    retq
-;
-; AVX512BW-LABEL: test_16f32toub:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512BW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512BW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512BW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $7, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $9, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $6, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $10, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $5, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $11, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $4, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $12, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $3, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $13, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $2, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftlw $1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k0
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    retq
+; ALL-LABEL: test_16f32toub:
+; ALL:       # %bb.0:
+; ALL-NEXT:    vcvttps2dq %zmm0, %zmm0
+; ALL-NEXT:    vpslld $31, %zmm0, %zmm0
+; ALL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; ALL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
+; ALL-NEXT:    retq
   %mask = fptoui <16 x float> %a to <16 x i1>
   %select = select <16 x i1> %mask, <16 x i32> %passthru, <16 x i32> zeroinitializer
   ret <16 x i32> %select
@@ -3878,43 +2226,13 @@ define <2 x i64> @test_2f64tosb(<2 x double> %a, <2 x i64> %passthru) {
 ; NOVLDQ-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; NOVLDQ-NEXT:    retq
 ;
-; VLBW-LABEL: test_2f64tosb:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    andl $1, %eax
-; VLBW-NEXT:    kmovw %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLBW-NEXT:    korw %k1, %k2, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k1
-; VLBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_2f64tosb:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    andl $1, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLNOBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLNOBW-NEXT:    korw %k1, %k2, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k1
-; VLNOBW-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
+; VL-LABEL: test_2f64tosb:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttpd2dq %xmm0, %xmm0
+; VL-NEXT:    vpslld $31, %xmm0, %xmm0
+; VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
+; VL-NEXT:    vmovdqa64 %xmm1, %xmm0 {%k1} {z}
+; VL-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: test_2f64tosb:
 ; AVX512DQ:       # %bb.0:
@@ -3936,534 +2254,31 @@ define <4 x i64> @test_4f64tosb(<4 x double> %a, <4 x i64> %passthru) {
 ; NOVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; NOVL-NEXT:    retq
 ;
-; VLBW-LABEL: test_4f64tosb:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    andl $1, %eax
-; VLBW-NEXT:    kmovw %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLBW-NEXT:    korw %k1, %k2, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k1
-; VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_4f64tosb:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttsd2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    andl $1, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLNOBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLNOBW-NEXT:    korw %k1, %k2, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLNOBW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttsd2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k1
-; VLNOBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
+; VL-LABEL: test_4f64tosb:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttpd2dq %ymm0, %xmm0
+; VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
+; VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptosi <4 x double> %a to <4 x i1>
   %select = select <4 x i1> %mask, <4 x i64> %passthru, <4 x i64> zeroinitializer
   ret <4 x i64> %select
 }
 
 define <8 x i64> @test_8f64tosb(<8 x double> %a, <8 x i64> %passthru) {
-; KNL-LABEL: test_8f64tosb:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k0
-; KNL-NEXT:    vcvttsd2si %xmm0, %eax
-; KNL-NEXT:    andl $1, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftrw $1, %k0, %k2
-; KNL-NEXT:    kshiftlw $1, %k2, %k2
-; KNL-NEXT:    korw %k1, %k2, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k2
-; KNL-NEXT:    kxorw %k0, %k2, %k0
-; KNL-NEXT:    kshiftlw $15, %k0, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k0
-; KNL-NEXT:    kxorw %k1, %k0, %k0
-; KNL-NEXT:    kshiftrw $2, %k0, %k1
-; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $13, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $3, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $12, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $4, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $11, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $5, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $10, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $6, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; KNL-NEXT:    vcvttsd2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $9, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $7, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; KNL-NEXT:    vcvttsd2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $8, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k1
-; KNL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; KNL-NEXT:    retq
-;
-; SKX-LABEL: test_8f64tosb:
-; SKX:       # %bb.0:
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k0
-; SKX-NEXT:    vcvttsd2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k2
-; SKX-NEXT:    kxorb %k0, %k2, %k0
-; SKX-NEXT:    kshiftlb $7, %k0, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k0
-; SKX-NEXT:    kxorb %k1, %k0, %k0
-; SKX-NEXT:    kshiftrb $2, %k0, %k1
-; SKX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $5, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $3, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $4, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $4, %k0, %k1
-; SKX-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $3, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $5, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $2, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k1
-; SKX-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; SKX-NEXT:    vcvttsd2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftlb $1, %k0, %k0
-; SKX-NEXT:    kshiftrb $1, %k0, %k0
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; SKX-NEXT:    vcvttsd2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    korb %k1, %k0, %k1
-; SKX-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; SKX-NEXT:    retq
-;
-; AVX512VL-LABEL: test_8f64tosb:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k0
-; AVX512VL-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VL-NEXT:    andl $1, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VL-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VL-NEXT:    korw %k1, %k2, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VL-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VL-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VL-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512VL-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VL-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_8f64tosb:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k0
-; AVX512DQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512DQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512DQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512DQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512DQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    korb %k1, %k0, %k1
-; AVX512DQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512DQ-NEXT:    retq
-;
-; AVX512BW-LABEL: test_8f64tosb:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k0
-; AVX512BW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512BW-NEXT:    andl $1, %eax
-; AVX512BW-NEXT:    kmovw %eax, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512BW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512BW-NEXT:    korw %k1, %k2, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512BW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512BW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512BW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    retq
-;
-; AVX512VLDQ-LABEL: test_8f64tosb:
-; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k0
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512VLDQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512VLDQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    korb %k1, %k0, %k1
-; AVX512VLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLDQ-NEXT:    retq
+; NOVL-LABEL: test_8f64tosb:
+; NOVL:       # %bb.0:
+; NOVL-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; NOVL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; NOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; NOVL-NEXT:    retq
 ;
-; AVX512VLBW-LABEL: test_8f64tosb:
-; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k0
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLBW-NEXT:    andl $1, %eax
-; AVX512VLBW-NEXT:    kmovw %eax, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VLBW-NEXT:    korw %k1, %k2, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VLBW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VLBW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm2[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttsd2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLBW-NEXT:    retq
+; VL-LABEL: test_8f64tosb:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttpd2dq %zmm0, %ymm0
+; VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
+; VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptosi <8 x double> %a to <8 x i1>
   %select = select <8 x i1> %mask, <8 x i64> %passthru, <8 x i64> zeroinitializer
   ret <8 x i64> %select
@@ -4531,1169 +2346,43 @@ define <4 x i64> @test_4f32tosb(<4 x float> %a, <4 x i64> %passthru) {
 ; NOVL-NEXT:    vpand %ymm1, %ymm0, %ymm0
 ; NOVL-NEXT:    retq
 ;
-; VLBW-LABEL: test_4f32tosb:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    andl $1, %eax
-; VLBW-NEXT:    kmovw %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLBW-NEXT:    korw %k1, %k2, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k1
-; VLBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_4f32tosb:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    andl $1, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k2
-; VLNOBW-NEXT:    kshiftlw $1, %k2, %k2
-; VLNOBW-NEXT:    korw %k1, %k2, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k1
-; VLNOBW-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
+; VL-LABEL: test_4f32tosb:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttps2dq %xmm0, %xmm0
+; VL-NEXT:    vptestmd %xmm0, %xmm0, %k1
+; VL-NEXT:    vmovdqa64 %ymm1, %ymm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptosi <4 x float> %a to <4 x i1>
   %select = select <4 x i1> %mask, <4 x i64> %passthru, <4 x i64> zeroinitializer
   ret <4 x i64> %select
 }
 
 define <8 x i64> @test_8f32tosb(<8 x float> %a, <8 x i64> %passthru) {
-; KNL-LABEL: test_8f32tosb:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    andl $1, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftrw $1, %k0, %k2
-; KNL-NEXT:    kshiftlw $1, %k2, %k2
-; KNL-NEXT:    korw %k1, %k2, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k2
-; KNL-NEXT:    kxorw %k0, %k2, %k0
-; KNL-NEXT:    kshiftlw $15, %k0, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k0
-; KNL-NEXT:    kxorw %k1, %k0, %k0
-; KNL-NEXT:    kshiftrw $2, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $13, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $3, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $12, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $4, %k0, %k1
-; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $11, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $5, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $10, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $6, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $9, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $7, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $8, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k1
-; KNL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; KNL-NEXT:    retq
-;
-; SKX-LABEL: test_8f32tosb:
-; SKX:       # %bb.0:
-; SKX-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k0
-; SKX-NEXT:    vcvttss2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k2
-; SKX-NEXT:    kxorb %k0, %k2, %k0
-; SKX-NEXT:    kshiftlb $7, %k0, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k0
-; SKX-NEXT:    kxorb %k1, %k0, %k0
-; SKX-NEXT:    kshiftrb $2, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $5, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $3, %k0, %k1
-; SKX-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $4, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $4, %k0, %k1
-; SKX-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; SKX-NEXT:    vcvttss2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $3, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $5, %k0, %k1
-; SKX-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $2, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftrb $6, %k0, %k1
-; SKX-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; SKX-NEXT:    vcvttss2si %xmm2, %eax
-; SKX-NEXT:    kmovd %eax, %k2
-; SKX-NEXT:    kxorb %k2, %k1, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    kshiftrb $1, %k1, %k1
-; SKX-NEXT:    kxorb %k0, %k1, %k0
-; SKX-NEXT:    kshiftlb $1, %k0, %k0
-; SKX-NEXT:    kshiftrb $1, %k0, %k0
-; SKX-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; SKX-NEXT:    vcvttss2si %xmm0, %eax
-; SKX-NEXT:    kmovd %eax, %k1
-; SKX-NEXT:    kshiftlb $7, %k1, %k1
-; SKX-NEXT:    korb %k1, %k0, %k1
-; SKX-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; SKX-NEXT:    retq
-;
-; AVX512VL-LABEL: test_8f32tosb:
-; AVX512VL:       # %bb.0:
-; AVX512VL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k0
-; AVX512VL-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VL-NEXT:    andl $1, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VL-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VL-NEXT:    korw %k1, %k2, %k1
-; AVX512VL-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VL-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VL-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VL-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VL-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VL-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512VL-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VL-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VL-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VL-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VL-NEXT:    kmovw %eax, %k2
-; AVX512VL-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VL-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VL-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VL-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VL-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_8f32tosb:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512DQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512DQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512DQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512DQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512DQ-NEXT:    korb %k1, %k0, %k1
-; AVX512DQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512DQ-NEXT:    retq
-;
-; AVX512BW-LABEL: test_8f32tosb:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    andl $1, %eax
-; AVX512BW-NEXT:    kmovw %eax, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512BW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512BW-NEXT:    korw %k1, %k2, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512BW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512BW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512BW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512BW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    retq
-;
-; AVX512VLDQ-LABEL: test_8f32tosb:
-; AVX512VLDQ:       # %bb.0:
-; AVX512VLDQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k0
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k2
-; AVX512VLDQ-NEXT:    kxorb %k0, %k2, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k0
-; AVX512VLDQ-NEXT:    kxorb %k1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $4, %k0, %k1
-; AVX512VLDQ-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $3, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $5, %k0, %k1
-; AVX512VLDQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $2, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $6, %k0, %k1
-; AVX512VLDQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k2
-; AVX512VLDQ-NEXT:    kxorb %k2, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k1, %k1
-; AVX512VLDQ-NEXT:    kxorb %k0, %k1, %k0
-; AVX512VLDQ-NEXT:    kshiftlb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    kshiftrb $1, %k0, %k0
-; AVX512VLDQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VLDQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLDQ-NEXT:    kmovw %eax, %k1
-; AVX512VLDQ-NEXT:    kshiftlb $7, %k1, %k1
-; AVX512VLDQ-NEXT:    korb %k1, %k0, %k1
-; AVX512VLDQ-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLDQ-NEXT:    retq
+; NOVL-LABEL: test_8f32tosb:
+; NOVL:       # %bb.0:
+; NOVL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; NOVL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; NOVL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; NOVL-NEXT:    retq
 ;
-; AVX512VLBW-LABEL: test_8f32tosb:
-; AVX512VLBW:       # %bb.0:
-; AVX512VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k0
-; AVX512VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLBW-NEXT:    andl $1, %eax
-; AVX512VLBW-NEXT:    kmovw %eax, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k0, %k2
-; AVX512VLBW-NEXT:    kshiftlw $1, %k2, %k2
-; AVX512VLBW-NEXT:    korw %k1, %k2, %k1
-; AVX512VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512VLBW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512VLBW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX512VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512VLBW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512VLBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512VLBW-NEXT:    kmovd %eax, %k2
-; AVX512VLBW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512VLBW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512VLBW-NEXT:    kxorw %k0, %k1, %k1
-; AVX512VLBW-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
-; AVX512VLBW-NEXT:    retq
+; VL-LABEL: test_8f32tosb:
+; VL:       # %bb.0:
+; VL-NEXT:    vcvttps2dq %ymm0, %ymm0
+; VL-NEXT:    vptestmd %ymm0, %ymm0, %k1
+; VL-NEXT:    vmovdqa64 %zmm1, %zmm0 {%k1} {z}
+; VL-NEXT:    retq
   %mask = fptosi <8 x float> %a to <8 x i1>
   %select = select <8 x i1> %mask, <8 x i64> %passthru, <8 x i64> zeroinitializer
   ret <8 x i64> %select
 }
 
 define <16 x i32> @test_16f32tosb(<16 x float> %a, <16 x i32> %passthru) {
-; KNL-LABEL: test_16f32tosb:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k2
-; KNL-NEXT:    kxorw %k0, %k2, %k0
-; KNL-NEXT:    kshiftlw $15, %k0, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k0
-; KNL-NEXT:    kxorw %k1, %k0, %k0
-; KNL-NEXT:    kshiftrw $2, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $13, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $3, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $12, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $4, %k0, %k1
-; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $11, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $5, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $10, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $6, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $9, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $7, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $8, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $8, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $7, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $9, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $6, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $10, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; KNL-NEXT:    vcvttss2si %xmm3, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $5, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $11, %k0, %k1
-; KNL-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $4, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $12, %k0, %k1
-; KNL-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $3, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $13, %k0, %k1
-; KNL-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $2, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftrw $14, %k0, %k1
-; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; KNL-NEXT:    vcvttss2si %xmm2, %eax
-; KNL-NEXT:    kmovw %eax, %k2
-; KNL-NEXT:    kxorw %k2, %k1, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    kshiftrw $1, %k1, %k1
-; KNL-NEXT:    kxorw %k0, %k1, %k0
-; KNL-NEXT:    kshiftlw $1, %k0, %k0
-; KNL-NEXT:    kshiftrw $1, %k0, %k0
-; KNL-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; KNL-NEXT:    vcvttss2si %xmm0, %eax
-; KNL-NEXT:    kmovw %eax, %k1
-; KNL-NEXT:    kshiftlw $15, %k1, %k1
-; KNL-NEXT:    korw %k1, %k0, %k1
-; KNL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; KNL-NEXT:    retq
-;
-; VLBW-LABEL: test_16f32tosb:
-; VLBW:       # %bb.0:
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k0
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLBW-NEXT:    kxorw %k0, %k2, %k0
-; VLBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLBW-NEXT:    kxorw %k1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $4, %k0, %k1
-; VLBW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $11, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $5, %k0, %k1
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $10, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $6, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $9, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $7, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $8, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $8, %k0, %k1
-; VLBW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $7, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $9, %k0, %k1
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $6, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $10, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $5, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $11, %k0, %k1
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $4, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $12, %k0, %k1
-; VLBW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $3, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $13, %k0, %k1
-; VLBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $2, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftrw $14, %k0, %k1
-; VLBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLBW-NEXT:    kmovd %eax, %k2
-; VLBW-NEXT:    kxorw %k2, %k1, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    kshiftrw $1, %k1, %k1
-; VLBW-NEXT:    kxorw %k0, %k1, %k0
-; VLBW-NEXT:    kshiftlw $1, %k0, %k0
-; VLBW-NEXT:    kshiftrw $1, %k0, %k0
-; VLBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLBW-NEXT:    kmovd %eax, %k1
-; VLBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLBW-NEXT:    korw %k1, %k0, %k1
-; VLBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; VLBW-NEXT:    retq
-;
-; VLNOBW-LABEL: test_16f32tosb:
-; VLNOBW:       # %bb.0:
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k0
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k2
-; VLNOBW-NEXT:    kxorw %k0, %k2, %k0
-; VLNOBW-NEXT:    kshiftlw $15, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k0
-; VLNOBW-NEXT:    kxorw %k1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $2, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $13, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $3, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $12, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $4, %k0, %k1
-; VLNOBW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $11, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $5, %k0, %k1
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $10, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $6, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $9, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $7, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $8, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $8, %k0, %k1
-; VLNOBW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $7, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $9, %k0, %k1
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $6, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $10, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm3, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $5, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $11, %k0, %k1
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $4, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $12, %k0, %k1
-; VLNOBW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $3, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $13, %k0, %k1
-; VLNOBW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $2, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftrw $14, %k0, %k1
-; VLNOBW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; VLNOBW-NEXT:    vcvttss2si %xmm2, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k2
-; VLNOBW-NEXT:    kxorw %k2, %k1, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    kshiftrw $1, %k1, %k1
-; VLNOBW-NEXT:    kxorw %k0, %k1, %k0
-; VLNOBW-NEXT:    kshiftlw $1, %k0, %k0
-; VLNOBW-NEXT:    kshiftrw $1, %k0, %k0
-; VLNOBW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; VLNOBW-NEXT:    vcvttss2si %xmm0, %eax
-; VLNOBW-NEXT:    kmovw %eax, %k1
-; VLNOBW-NEXT:    kshiftlw $15, %k1, %k1
-; VLNOBW-NEXT:    korw %k1, %k0, %k1
-; VLNOBW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; VLNOBW-NEXT:    retq
-;
-; AVX512DQ-LABEL: test_16f32tosb:
-; AVX512DQ:       # %bb.0:
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512DQ-NEXT:    kxorw %k0, %k2, %k0
-; AVX512DQ-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512DQ-NEXT:    kxorw %k1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512DQ-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $7, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $9, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $6, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $10, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $5, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $11, %k0, %k1
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $4, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $12, %k0, %k1
-; AVX512DQ-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $3, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $13, %k0, %k1
-; AVX512DQ-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $2, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftrw $14, %k0, %k1
-; AVX512DQ-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512DQ-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k2
-; AVX512DQ-NEXT:    kxorw %k2, %k1, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    kshiftrw $1, %k1, %k1
-; AVX512DQ-NEXT:    kxorw %k0, %k1, %k0
-; AVX512DQ-NEXT:    kshiftlw $1, %k0, %k0
-; AVX512DQ-NEXT:    kshiftrw $1, %k0, %k0
-; AVX512DQ-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512DQ-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512DQ-NEXT:    kmovw %eax, %k1
-; AVX512DQ-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512DQ-NEXT:    korw %k1, %k0, %k1
-; AVX512DQ-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; AVX512DQ-NEXT:    retq
-;
-; AVX512BW-LABEL: test_16f32tosb:
-; AVX512BW:       # %bb.0:
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k2
-; AVX512BW-NEXT:    kxorw %k0, %k2, %k0
-; AVX512BW-NEXT:    kshiftlw $15, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k0
-; AVX512BW-NEXT:    kxorw %k1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $2, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $13, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $3, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $12, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $4, %k0, %k1
-; AVX512BW-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $11, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $5, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $10, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $6, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $9, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $7, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $8, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $8, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $2, %zmm0, %xmm2
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $7, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $9, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm3 = xmm2[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $6, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $10, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm2[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm3, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $5, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $11, %k0, %k1
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm2 = xmm2[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $4, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $12, %k0, %k1
-; AVX512BW-NEXT:    vextractf32x4 $3, %zmm0, %xmm0
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $3, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $13, %k0, %k1
-; AVX512BW-NEXT:    vmovshdup {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $2, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftrw $14, %k0, %k1
-; AVX512BW-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
-; AVX512BW-NEXT:    vcvttss2si %xmm2, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k2
-; AVX512BW-NEXT:    kxorw %k2, %k1, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    kshiftrw $1, %k1, %k1
-; AVX512BW-NEXT:    kxorw %k0, %k1, %k0
-; AVX512BW-NEXT:    kshiftlw $1, %k0, %k0
-; AVX512BW-NEXT:    kshiftrw $1, %k0, %k0
-; AVX512BW-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[3,1,2,3]
-; AVX512BW-NEXT:    vcvttss2si %xmm0, %eax
-; AVX512BW-NEXT:    kmovd %eax, %k1
-; AVX512BW-NEXT:    kshiftlw $15, %k1, %k1
-; AVX512BW-NEXT:    korw %k1, %k0, %k1
-; AVX512BW-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
-; AVX512BW-NEXT:    retq
+; ALL-LABEL: test_16f32tosb:
+; ALL:       # %bb.0:
+; ALL-NEXT:    vcvttps2dq %zmm0, %zmm0
+; ALL-NEXT:    vptestmd %zmm0, %zmm0, %k1
+; ALL-NEXT:    vmovdqa32 %zmm1, %zmm0 {%k1} {z}
+; ALL-NEXT:    retq
   %mask = fptosi <16 x float> %a to <16 x i1>
   %select = select <16 x i1> %mask, <16 x i32> %passthru, <16 x i32> zeroinitializer
   ret <16 x i32> %select