From 8b69549dc5c5fa0f5f8632cde1c740bb2c7d8957 Mon Sep 17 00:00:00 2001
From: Phoebe Wang <phoebe.wang@intel.com>
Date: Sun, 14 Aug 2022 09:03:09 +0800
Subject: [PATCH] [X86][FP16] Promote FP16->[U]INT to FP16->FP32->[U]INT

This is to avoid f16->i64 being lowered to `__fixhfdi/__fixunshfdi` on 32-bits since neither libgcc nor compiler-rt provide them. https://godbolt.org/z/cjWEsea5v

It also helps to improve the performance by promoting the vector type.

Reviewed By: LuoYuanke

Differential Revision: https://reviews.llvm.org/D131828
---
 llvm/lib/Target/X86/X86ISelLowering.cpp       |  23 +-
 llvm/test/CodeGen/X86/fpclamptosat_vec.ll     | 348 ++++++++----------
 llvm/test/CodeGen/X86/half.ll                 |  46 ++-
 .../CodeGen/X86/vector-half-conversions.ll    |  55 ++-
 4 files changed, 246 insertions(+), 226 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index d5d41ca50553..3ff51374c114 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -32734,8 +32734,29 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
                     N->getOpcode() == ISD::STRICT_FP_TO_SINT;
     EVT VT = N->getValueType(0);
     SDValue Src = N->getOperand(IsStrict ? 1 : 0);
+    SDValue Chain = IsStrict ? N->getOperand(0) : SDValue();
     EVT SrcVT = Src.getValueType();
 
+    SDValue Res;
+    if (isSoftFP16(SrcVT)) {
+      EVT NVT = VT.isVector() ? VT.changeVectorElementType(MVT::f32) : MVT::f32;
+      if (IsStrict) {
+        Res =
+            DAG.getNode(N->getOpcode(), dl, {VT, MVT::Other},
+                        {Chain, DAG.getNode(ISD::STRICT_FP_EXTEND, dl,
+                                            {NVT, MVT::Other}, {Chain, Src})});
+        Chain = Res.getValue(1);
+      } else {
+        Res = DAG.getNode(N->getOpcode(), dl, VT,
+                          DAG.getNode(ISD::FP_EXTEND, dl, NVT, Src));
+      }
+      Results.push_back(Res);
+      if (IsStrict)
+        Results.push_back(Chain);
+
+      return;
+    }
+
     if (VT.isVector() && Subtarget.hasFP16() &&
         SrcVT.getVectorElementType() == MVT::f16) {
       EVT EleVT = VT.getVectorElementType();
@@ -32749,7 +32770,6 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
         Src = DAG.getNode(ISD::CONCAT_VECTORS, dl, MVT::v8f16, Ops);
       }
 
-      SDValue Res, Chain;
       if (IsStrict) {
         unsigned Opc =
             IsSigned ? X86ISD::STRICT_CVTTP2SI : X86ISD::STRICT_CVTTP2UI;
@@ -32941,7 +32961,6 @@ void X86TargetLowering::ReplaceNodeResults(SDNode *N,
       return;
     }
 
-    SDValue Chain;
     if (SDValue V = FP_TO_INTHelper(SDValue(N, 0), DAG, IsSigned, Chain)) {
       Results.push_back(V);
       if (IsStrict)
diff --git a/llvm/test/CodeGen/X86/fpclamptosat_vec.ll b/llvm/test/CodeGen/X86/fpclamptosat_vec.ll
index c22656dc2a16..c6883afe07ed 100644
--- a/llvm/test/CodeGen/X86/fpclamptosat_vec.ll
+++ b/llvm/test/CodeGen/X86/fpclamptosat_vec.ll
@@ -863,59 +863,47 @@ define <8 x i16> @stest_f16i16(<8 x half> %x) {
 ; CHECK-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrlq $48, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrld $16, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm1
-; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; CHECK-NEXT:    cvttps2dq %xmm1, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
 ; CHECK-NEXT:    packssdw {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
@@ -940,82 +928,94 @@ define <8 x i16> @utesth_f16i16(<8 x half> %x) {
 ; CHECK-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrad $31, %xmm2
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrad $31, %xmm2
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrlq $48, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrad $31, %xmm2
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrld $16, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1]
-; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm4 = xmm4[0],mem[0]
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; CHECK-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
+; CHECK-NEXT:    cvttps2dq %xmm2, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrad $31, %xmm1
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; CHECK-NEXT:    cvttps2dq %xmm2, %xmm2
+; CHECK-NEXT:    pand %xmm1, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
+; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm2 = xmm2[0],mem[0]
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [2147483648,2147483648,2147483648,2147483648]
-; CHECK-NEXT:    movdqa %xmm4, %xmm2
-; CHECK-NEXT:    pxor %xmm1, %xmm2
-; CHECK-NEXT:    movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; CHECK-NEXT:    movdqa %xmm3, %xmm0
-; CHECK-NEXT:    pcmpgtd %xmm2, %xmm0
-; CHECK-NEXT:    pand %xmm0, %xmm4
-; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
-; CHECK-NEXT:    pxor %xmm2, %xmm0
-; CHECK-NEXT:    por %xmm4, %xmm0
-; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; CHECK-NEXT:    pxor %xmm4, %xmm1
-; CHECK-NEXT:    pcmpgtd %xmm1, %xmm3
-; CHECK-NEXT:    pand %xmm3, %xmm4
-; CHECK-NEXT:    pxor %xmm2, %xmm3
-; CHECK-NEXT:    por %xmm4, %xmm3
-; CHECK-NEXT:    pslld $16, %xmm3
-; CHECK-NEXT:    psrad $16, %xmm3
+; CHECK-NEXT:    movdqa %xmm2, %xmm3
+; CHECK-NEXT:    pxor %xmm1, %xmm3
+; CHECK-NEXT:    movdqa {{.*#+}} xmm4 = [2147549183,2147549183,2147549183,2147549183]
+; CHECK-NEXT:    movdqa %xmm4, %xmm0
+; CHECK-NEXT:    pcmpgtd %xmm3, %xmm0
+; CHECK-NEXT:    pand %xmm0, %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm3, %xmm3
+; CHECK-NEXT:    pxor %xmm3, %xmm0
+; CHECK-NEXT:    por %xmm2, %xmm0
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; CHECK-NEXT:    pxor %xmm2, %xmm1
+; CHECK-NEXT:    pcmpgtd %xmm1, %xmm4
+; CHECK-NEXT:    pand %xmm4, %xmm2
+; CHECK-NEXT:    pxor %xmm3, %xmm4
+; CHECK-NEXT:    por %xmm2, %xmm4
+; CHECK-NEXT:    pslld $16, %xmm4
+; CHECK-NEXT:    psrad $16, %xmm4
 ; CHECK-NEXT:    pslld $16, %xmm0
 ; CHECK-NEXT:    psrad $16, %xmm0
-; CHECK-NEXT:    packssdw %xmm3, %xmm0
+; CHECK-NEXT:    packssdw %xmm4, %xmm0
 ; CHECK-NEXT:    addq $72, %rsp
 ; CHECK-NEXT:    .cfi_def_cfa_offset 8
 ; CHECK-NEXT:    retq
@@ -1035,59 +1035,47 @@ define <8 x i16> @ustest_f16i16(<8 x half> %x) {
 ; CHECK-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
 ; CHECK-NEXT:    psrlq $48, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrld $16, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
 ; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm1 = xmm1[0],mem[0]
-; CHECK-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm1, %xmm0
+; CHECK-NEXT:    unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
@@ -2437,59 +2425,47 @@ define <8 x i16> @stest_f16i16_mm(<8 x half> %x) {
 ; CHECK-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrlq $48, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrld $16, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm1
-; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
+; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
+; CHECK-NEXT:    cvttps2dq %xmm1, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
 ; CHECK-NEXT:    packssdw {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
@@ -2512,82 +2488,94 @@ define <8 x i16> @utesth_f16i16_mm(<8 x half> %x) {
 ; CHECK-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrad $31, %xmm2
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrad $31, %xmm2
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrlq $48, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm1
+; CHECK-NEXT:    movdqa %xmm1, %xmm2
+; CHECK-NEXT:    psrad $31, %xmm2
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    pand %xmm2, %xmm0
+; CHECK-NEXT:    por %xmm1, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrld $16, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %rax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; CHECK-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm0[0],xmm4[1],xmm0[1]
-; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm4 = xmm4[0],mem[0]
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; CHECK-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1]
+; CHECK-NEXT:    cvttps2dq %xmm2, %xmm0
+; CHECK-NEXT:    movdqa %xmm0, %xmm1
+; CHECK-NEXT:    psrad $31, %xmm1
+; CHECK-NEXT:    subps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2
+; CHECK-NEXT:    cvttps2dq %xmm2, %xmm2
+; CHECK-NEXT:    pand %xmm1, %xmm2
+; CHECK-NEXT:    por %xmm0, %xmm2
+; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm2 = xmm2[0],mem[0]
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [2147483648,2147483648,2147483648,2147483648]
-; CHECK-NEXT:    movdqa %xmm4, %xmm2
-; CHECK-NEXT:    pxor %xmm1, %xmm2
-; CHECK-NEXT:    movdqa {{.*#+}} xmm3 = [2147549183,2147549183,2147549183,2147549183]
-; CHECK-NEXT:    movdqa %xmm3, %xmm0
-; CHECK-NEXT:    pcmpgtd %xmm2, %xmm0
-; CHECK-NEXT:    pand %xmm0, %xmm4
-; CHECK-NEXT:    pcmpeqd %xmm2, %xmm2
-; CHECK-NEXT:    pxor %xmm2, %xmm0
-; CHECK-NEXT:    por %xmm4, %xmm0
-; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm4 # 16-byte Reload
-; CHECK-NEXT:    pxor %xmm4, %xmm1
-; CHECK-NEXT:    pcmpgtd %xmm1, %xmm3
-; CHECK-NEXT:    pand %xmm3, %xmm4
-; CHECK-NEXT:    pxor %xmm2, %xmm3
-; CHECK-NEXT:    por %xmm4, %xmm3
-; CHECK-NEXT:    pslld $16, %xmm3
-; CHECK-NEXT:    psrad $16, %xmm3
+; CHECK-NEXT:    movdqa %xmm2, %xmm3
+; CHECK-NEXT:    pxor %xmm1, %xmm3
+; CHECK-NEXT:    movdqa {{.*#+}} xmm4 = [2147549183,2147549183,2147549183,2147549183]
+; CHECK-NEXT:    movdqa %xmm4, %xmm0
+; CHECK-NEXT:    pcmpgtd %xmm3, %xmm0
+; CHECK-NEXT:    pand %xmm0, %xmm2
+; CHECK-NEXT:    pcmpeqd %xmm3, %xmm3
+; CHECK-NEXT:    pxor %xmm3, %xmm0
+; CHECK-NEXT:    por %xmm2, %xmm0
+; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
+; CHECK-NEXT:    pxor %xmm2, %xmm1
+; CHECK-NEXT:    pcmpgtd %xmm1, %xmm4
+; CHECK-NEXT:    pand %xmm4, %xmm2
+; CHECK-NEXT:    pxor %xmm3, %xmm4
+; CHECK-NEXT:    por %xmm2, %xmm4
+; CHECK-NEXT:    pslld $16, %xmm4
+; CHECK-NEXT:    psrad $16, %xmm4
 ; CHECK-NEXT:    pslld $16, %xmm0
 ; CHECK-NEXT:    psrad $16, %xmm0
-; CHECK-NEXT:    packssdw %xmm3, %xmm0
+; CHECK-NEXT:    packssdw %xmm4, %xmm0
 ; CHECK-NEXT:    addq $72, %rsp
 ; CHECK-NEXT:    .cfi_def_cfa_offset 8
 ; CHECK-NEXT:    retq
@@ -2606,59 +2594,47 @@ define <8 x i16> @ustest_f16i16_mm(<8 x half> %x) {
 ; CHECK-NEXT:    movdqa %xmm0, (%rsp) # 16-byte Spill
 ; CHECK-NEXT:    psrlq $48, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrld $16, %xmm0
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
 ; CHECK-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload
-; CHECK-NEXT:    # xmm1 = xmm1[0],mem[0]
-; CHECK-NEXT:    movdqa %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm1, %xmm0
+; CHECK-NEXT:    unpcklpd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
-; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
+; CHECK-NEXT:    movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movdqa (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    psrldq {{.*#+}} xmm0 = xmm0[10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
 ; CHECK-NEXT:    movaps (%rsp), %xmm0 # 16-byte Reload
 ; CHECK-NEXT:    movhlps {{.*#+}} xmm0 = xmm0[1,1]
 ; CHECK-NEXT:    callq __extendhfsf2@PLT
-; CHECK-NEXT:    cvttss2si %xmm0, %eax
-; CHECK-NEXT:    movd %eax, %xmm0
-; CHECK-NEXT:    punpckldq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT:    unpcklps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
+; CHECK-NEXT:    cvttps2dq %xmm0, %xmm0
 ; CHECK-NEXT:    punpcklqdq {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
 ; CHECK-NEXT:    # xmm0 = xmm0[0],mem[0]
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
diff --git a/llvm/test/CodeGen/X86/half.ll b/llvm/test/CodeGen/X86/half.ll
index 7ec40d8b220e..501db6b2dba0 100644
--- a/llvm/test/CodeGen/X86/half.ll
+++ b/llvm/test/CodeGen/X86/half.ll
@@ -228,13 +228,24 @@ define i64 @test_fptosi_i64(ptr %p) #0 {
 ;
 ; CHECK-I686-LABEL: test_fptosi_i64:
 ; CHECK-I686:       # %bb.0:
-; CHECK-I686-NEXT:    subl $12, %esp
+; CHECK-I686-NEXT:    subl $28, %esp
 ; CHECK-I686-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-I686-NEXT:    pinsrw $0, (%eax), %xmm0
 ; CHECK-I686-NEXT:    pextrw $0, %xmm0, %eax
 ; CHECK-I686-NEXT:    movw %ax, (%esp)
-; CHECK-I686-NEXT:    calll __fixhfdi
-; CHECK-I686-NEXT:    addl $12, %esp
+; CHECK-I686-NEXT:    calll __extendhfsf2
+; CHECK-I686-NEXT:    fstps {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fnstcw {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; CHECK-I686-NEXT:    orl $3072, %eax # imm = 0xC00
+; CHECK-I686-NEXT:    movw %ax, {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fldcw {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fistpll {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fldcw {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; CHECK-I686-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; CHECK-I686-NEXT:    addl $28, %esp
 ; CHECK-I686-NEXT:    retl
   %a = load half, ptr %p, align 2
   %r = fptosi half %a to i64
@@ -315,13 +326,36 @@ define i64 @test_fptoui_i64(ptr %p) #0 {
 ;
 ; CHECK-I686-LABEL: test_fptoui_i64:
 ; CHECK-I686:       # %bb.0:
-; CHECK-I686-NEXT:    subl $12, %esp
+; CHECK-I686-NEXT:    subl $28, %esp
 ; CHECK-I686-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-I686-NEXT:    pinsrw $0, (%eax), %xmm0
 ; CHECK-I686-NEXT:    pextrw $0, %xmm0, %eax
 ; CHECK-I686-NEXT:    movw %ax, (%esp)
-; CHECK-I686-NEXT:    calll __fixunshfdi
-; CHECK-I686-NEXT:    addl $12, %esp
+; CHECK-I686-NEXT:    calll __extendhfsf2
+; CHECK-I686-NEXT:    fstps {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-I686-NEXT:    movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-I686-NEXT:    ucomiss %xmm1, %xmm0
+; CHECK-I686-NEXT:    jae .LBB9_2
+; CHECK-I686-NEXT:  # %bb.1:
+; CHECK-I686-NEXT:    xorps %xmm1, %xmm1
+; CHECK-I686-NEXT:  .LBB9_2:
+; CHECK-I686-NEXT:    subss %xmm1, %xmm0
+; CHECK-I686-NEXT:    movss %xmm0, {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    setae %al
+; CHECK-I686-NEXT:    flds {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fnstcw {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    movzwl {{[0-9]+}}(%esp), %ecx
+; CHECK-I686-NEXT:    orl $3072, %ecx # imm = 0xC00
+; CHECK-I686-NEXT:    movw %cx, {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fldcw {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fistpll {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    fldcw {{[0-9]+}}(%esp)
+; CHECK-I686-NEXT:    movzbl %al, %edx
+; CHECK-I686-NEXT:    shll $31, %edx
+; CHECK-I686-NEXT:    xorl {{[0-9]+}}(%esp), %edx
+; CHECK-I686-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; CHECK-I686-NEXT:    addl $28, %esp
 ; CHECK-I686-NEXT:    retl
   %a = load half, ptr %p, align 2
   %r = fptoui half %a to i64
diff --git a/llvm/test/CodeGen/X86/vector-half-conversions.ll b/llvm/test/CodeGen/X86/vector-half-conversions.ll
index ac14b415f35b..ec9bb73bd2d4 100644
--- a/llvm/test/CodeGen/X86/vector-half-conversions.ll
+++ b/llvm/test/CodeGen/X86/vector-half-conversions.ll
@@ -1271,27 +1271,3 @@ define void @store_cvt_32f32_to_32f16(<32 x float> %a0, ptr %a1) nounwind {
   store <32 x half> %1, ptr %a1
   ret void
 }
-
-define <4 x i32> @fptosi_2f16_to_4i32(<2 x half> %a) nounwind {
-; ALL-LABEL: fptosi_2f16_to_4i32:
-; ALL:       # %bb.0:
-; ALL-NEXT:    vpsrld $16, %xmm0, %xmm1
-; ALL-NEXT:    vpextrw $0, %xmm1, %eax
-; ALL-NEXT:    movzwl %ax, %eax
-; ALL-NEXT:    vmovd %eax, %xmm1
-; ALL-NEXT:    vcvtph2ps %xmm1, %xmm1
-; ALL-NEXT:    vcvttss2si %xmm1, %eax
-; ALL-NEXT:    vpextrw $0, %xmm0, %ecx
-; ALL-NEXT:    movzwl %cx, %ecx
-; ALL-NEXT:    vmovd %ecx, %xmm0
-; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
-; ALL-NEXT:    vcvttss2si %xmm0, %ecx
-; ALL-NEXT:    vmovd %ecx, %xmm0
-; ALL-NEXT:    vmovd %eax, %xmm1
-; ALL-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; ALL-NEXT:    vmovq {{.*#+}} xmm0 = xmm0[0],zero
-; ALL-NEXT:    retq
-  %cvt = fptosi <2 x half> %a to <2 x i32>
-  %ext = shufflevector <2 x i32> %cvt, <2 x i32> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-  ret <4 x i32> %ext
-}
-- 
2.37.1.595.g718a3a8f04-goog