Index: lib/Target/AArch64/AArch64ISelLowering.cpp
===================================================================
--- lib/Target/AArch64/AArch64ISelLowering.cpp
+++ lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -6949,7 +6949,8 @@
   return SDValue();
 }
 
-static SDValue performIntToFpCombine(SDNode *N, SelectionDAG &DAG) {
+static SDValue performIntToFpCombine(SDNode *N, SelectionDAG &DAG,
+                                     const AArch64Subtarget *Subtarget) {
   // First try to optimize away the conversion when it's conditionally from
   // a constant. Vectors only.
   SDValue Res = performVectorCompareAndMaskUnaryOpCombine(N, DAG);
@@ -6968,7 +6969,7 @@
   // conversion, use a fp load instead and a AdvSIMD scalar {S|U}CVTF instead.
   // This eliminates an "integer-to-vector-move UOP and improve throughput.
   SDValue N0 = N->getOperand(0);
-  if (ISD::isNormalLoad(N0.getNode()) && N0.hasOneUse() &&
+  if (Subtarget->hasNEON() && ISD::isNormalLoad(N0.getNode()) && N0.hasOneUse() &&
       // Do not change the width of a volatile load.
       !cast<LoadSDNode>(N0)->isVolatile()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
@@ -8525,7 +8526,7 @@
     return performMulCombine(N, DAG, DCI, Subtarget);
   case ISD::SINT_TO_FP:
   case ISD::UINT_TO_FP:
-    return performIntToFpCombine(N, DAG);
+    return performIntToFpCombine(N, DAG, Subtarget);
   case ISD::OR:
     return performORCombine(N, DCI, Subtarget);
   case ISD::INTRINSIC_WO_CHAIN:
Index: test/CodeGen/AArch64/aarch64-2014-12-02-combine-soften.ll
===================================================================
--- /dev/null
+++ test/CodeGen/AArch64/aarch64-2014-12-02-combine-soften.ll
@@ -0,0 +1,16 @@
+;RUN: llc <%s -mattr=-neon  -mattr=-fp-armv8  | FileCheck %s
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64"
+
+@t = common global i32 0, align 4
+@x = common global i32 0, align 4
+
+define void @foo() {
+entry:
+;CHECK-LABEL: foo:
+;CHECK: __floatsisf
+  %0 = load i32* @x, align 4
+  %conv = sitofp i32 %0 to float
+  store float %conv, float* bitcast (i32* @t to float*), align 4
+  ret void
+}