Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -746,6 +746,11 @@
       // them legal.
       if (VT.getVectorElementType() == MVT::i1)
         setLoadExtAction(ISD::EXTLOAD, InnerVT, VT, Expand);
+
+      // EXTLOAD for MVT::f16 vectors is not legal because f16 vectors are
+      // split/scalarized right now.
+      if (VT.getVectorElementType() == MVT::f16)
+        setLoadExtAction(ISD::EXTLOAD, InnerVT, VT, Expand);
     }
   }
 
@@ -17505,6 +17510,11 @@
     return;
   }
   case ISD::FP_TO_SINT:
+    // FP_TO_INT*_IN_MEM is not legal for f16 inputs.  Do not convert
+    // (FP_TO_SINT (load f16)) to FP_TO_INT*.
+    if (N->getOperand(0).getValueType() == MVT::f16)
+      break;
+    // fallthrough in the false-branch
   case ISD::FP_TO_UINT: {
     bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT;
 
@@ -17550,6 +17560,13 @@
     Results.push_back(V);
     return;
   }
+  case ISD::FP_EXTEND: {
+    // Right now, only MVT::v2f32 has OperationAction for FP_EXTEND.
+    // No other ValueType for FP_EXTEND should reach this point.
+    assert(N->getValueType(0) == MVT::v2f32 &&
+           "Do not know how to legalize this Node");
+    return;
+  }
   case ISD::INTRINSIC_W_CHAIN: {
     unsigned IntNo = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
     switch (IntNo) {
@@ -23876,6 +23893,11 @@
   if (Op0.getOpcode() == ISD::LOAD) {
     LoadSDNode *Ld = cast<LoadSDNode>(Op0.getNode());
     EVT VT = Ld->getValueType(0);
+
+    // This transformation is not supported if the result type is f16
+    if (N->getValueType(0) == MVT::f16)
+      return SDValue();
+
     if (!Ld->isVolatile() && !N->getValueType(0).isVector() &&
         ISD::isNON_EXTLoad(Op0.getNode()) && Op0.hasOneUse() &&
         !Subtarget->is64Bit() && VT == MVT::i64) {
Index: test/CodeGen/X86/half.ll
===================================================================
--- test/CodeGen/X86/half.ll
+++ test/CodeGen/X86/half.ll
@@ -67,3 +67,166 @@
   store half %val16, half* %addr
   ret void
 }
+
+define i64 @test_fptosi_i64(half* %p) #0 {
+; CHECK-LABEL: test_fptosi_i64:
+
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-LIBCALL: cvttss2si
+; CHECK-FP16: vcvtph2ps
+; CHECK-FP16: vcvttss2si
+  %a = load half, half* %p, align 2
+  %r = fptosi half %a to i64
+  ret i64 %r
+}
+
+define void @test_sitofp_i64(i64 %a, half* %p) #0 {
+; CHECK-LABEL: test_sitofp_i64:
+
+; CHECK-LIBCALL: movq %rsi, [[ADDR:%[a-z]+]]
+; CHECK-LIBCALL: cvtsi2ssq
+; CHECK-LIBCALL: callq __gnu_f2h_ieee
+; CHECK-LIBCALL: movw [[SRC:%[a-z]+]], ([[ADDR]])
+; CHECK-FP16: vcvtsi2ssq
+; CHECK-FP16: vcvtps2ph
+; CHECK-FP16: movw [[SRC:%[a-z]+]], (%rsi)
+  %r = sitofp i64 %a to half
+  store half %r, half* %p
+  ret void
+}
+
+define i64 @test_fptoui_i64(half* %p) #0 {
+; CHECK-LABEL: test_fptoui_i64:
+
+; FP_TO_UINT is expanded using FP_TO_SINT
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-LIBCALL: movss
+; CHECK-LIBCALL: movaps
+; CHECK-LIBCALL: subss
+; CHECK-LIBCALL: cvttss2si
+; CHECK-LIBCALL: movabsq
+; CHECK-LIBCALL: xorq
+; CHECK-LIBCALL: cvttss2si
+; CHECK-LIBCALL: ucomiss
+; CHECK-LIBCALL: cmovaeq
+
+; CHECK-FP16: vcvtph2ps
+; CHECK-FP16: vmovss
+; CHECK-FP16: vmovaps
+; CHECK-FP16: vsubss
+; CHECK-FP16: vcvttss2si
+; CHECK-FP16: movabsq
+; CHECK-FP16: xorg
+; CHECK-FP16: vcvttss2si
+; CHECK-FP16: vucomiss
+; CHECK-FP16: cmovaeq
+  %a = load half, half* %p, align 2
+  %r = fptoui half %a to i64
+  ret i64 %r
+}
+
+define void @test_uitofp_i64(i64 %a, half* %p) #0 {
+; CHECK-LABEL: test_uitofp_i64:
+; CHECK: andl
+; CHECK-NEXT: testq %rdi, %rdi
+; CHECK-NEXT: js [[LABEL1:.LBB[0-9_]+]]
+
+; simple conversion to float if non-negative
+; CHECK-LIBCALL: cvtsi2ssq
+; CHECK-FP16: vcvtsi2ssq
+; CHECK: jmp [[LABEL2:.LBB[0-9_]+]]
+
+; convert using shift+or if negative
+; CHECK: [[LABEL1]]:
+; CHECK: shrq
+; CHECK-NEXT: orq
+; CHECK-LIBCALL: cvtsi2ssq
+; CHECK-LIBCALL: addss
+; CHECK-FP16: vcvtsi2ssq
+; CHECK-FP16: vaddss
+
+; convert float to half
+; CHECK: [[LABEL2]]:
+; CHECK-LIBCALL: callq __gnu_f2h_ieee
+; CHECK-FP16: vcvtps2ph
+; CHECK-FP16: movw [[SRC:%[a-z]+]], (%rsi)
+
+  %r = uitofp i64 %a to half
+  store half %r, half* %p
+  ret void
+}
+
+define <4 x float> @test_extend32_vec4(<4 x half>* %p) #0 {
+; CHECK-LABEL: test_extend32_vec4:
+
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-FP16: vcvtph2ps
+; CHECK-FP16: vcvtph2ps
+; CHECK-FP16: vcvtph2ps
+; CHECK-FP16: vcvtph2ps
+  %a = load <4 x half>, <4 x half>* %p, align 8
+  %b = fpext <4 x half> %a to <4 x float>
+  ret <4 x float> %b
+}
+
+define <4 x double> @test_extend64_vec4(<4 x half>* %p) #0 {
+; CHECK-LABEL: test_extend64_vec4
+
+; CHECK-LIBCALL: callq __gnu_h2f_ieee
+; CHECK-LIBCALL-DAG: callq __gnu_h2f_ieee
+; CHECK-LIBCALL-DAG: callq __gnu_h2f_ieee
+; CHECK-LIBCALL-DAG: callq __gnu_h2f_ieee
+; CHECK-LIBCALL-DAG: cvtss2sd
+; CHECK-LIBCALL-DAG: cvtss2sd
+; CHECK-LIBCALL-DAG: cvtss2sd
+; CHECK-LIBCALL: cvtss2sd
+; CHECK-FP16: vcvtph2ps
+; CHECK-FP16-DAG: vcvtph2ps
+; CHECK-FP16-DAG: vcvtph2ps
+; CHECK-FP16-DAG: vcvtph2ps
+; CHECK-FP16-DAG: vcvtss2sd
+; CHECK-FP16-DAG: vcvtss2sd
+; CHECK-FP16-DAG: vcvtss2sd
+; CHECK-FP16: vcvtss2sd
+  %a = load <4 x half>, <4 x half>* %p, align 8
+  %b = fpext <4 x half> %a to <4 x double>
+  ret <4 x double> %b
+}
+
+define void @test_trunc32_vec4(<4 x float> %a, <4 x half>* %p) {
+; CHECK-LABEL: test_trunc32_vec4:
+
+; CHECK-LIBCALL: callq __gnu_f2h_ieee
+; CHECK-LIBCALL: callq __gnu_f2h_ieee
+; CHECK-LIBCALL: callq __gnu_f2h_ieee
+; CHECK-LIBCALL: callq __gnu_f2h_ieee
+; CHECK-FP16: vcvtps2ph
+; CHECK-FP16: vcvtps2ph
+; CHECK-FP16: vcvtps2ph
+; CHECK-FP16: vcvtps2ph
+; CHECK: movw
+; CHECK: movw
+; CHECK: movw
+; CHECK: movw
+  %v = fptrunc <4 x float> %a to <4 x half>
+  store <4 x half> %v, <4 x half>* %p
+  ret void
+}
+
+define void @test_trunc64_vec4(<4 x double> %a, <4 x half>* %p) {
+; CHECK-LABEL: test_trunc64_vec4:
+; CHECK: callq  __truncdfhf2
+; CHECK: callq  __truncdfhf2
+; CHECK: callq  __truncdfhf2
+; CHECK: callq  __truncdfhf2
+; CHECK: movw
+; CHECK: movw
+; CHECK: movw
+; CHECK: movw
+  %v = fptrunc <4 x double> %a to <4 x half>
+  store <4 x half> %v, <4 x half>* %p
+  ret void
+}