Index: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
@@ -485,6 +485,9 @@
     setOperationAction(ISD::ATOMIC_STORE, VT, Custom);
   }
 
+  if (!Subtarget.is64Bit())
+    setOperationAction(ISD::ATOMIC_LOAD, MVT::i64, Custom);
+
   if (Subtarget.hasCmpxchg16b()) {
     setOperationAction(ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS, MVT::i128, Custom);
   }
@@ -25494,11 +25497,22 @@
 }
 
 // Note: this turns large loads into lock cmpxchg8b/16b.
-// FIXME: On 32 bits x86, fild/movq might be faster than lock cmpxchg8b.
+// TODO: In 32-bit mode, use MOVLPS when SSE1 is available?
+// TODO: In 32-bit mode, use FILD/FISTP when X87 is available?
 TargetLowering::AtomicExpansionKind
 X86TargetLowering::shouldExpandAtomicLoadInIR(LoadInst *LI) const {
-  return needsCmpXchgNb(LI->getType()) ? AtomicExpansionKind::CmpXChg
-                                       : AtomicExpansionKind::None;
+  Type *MemType = LI->getType();
+
+  // If this a 64 bit atomic load on a 32-bit target and SSE2 is enabled, we
+  // can use movq to do the load.
+  bool NoImplicitFloatOps =
+      LI->getFunction()->hasFnAttribute(Attribute::NoImplicitFloat);
+  if (MemType->getPrimitiveSizeInBits() == 64 && !Subtarget.is64Bit() &&
+      !Subtarget.useSoftFloat() && !NoImplicitFloatOps && Subtarget.hasSSE2())
+    return AtomicExpansionKind::None;
+
+  return needsCmpXchgNb(MemType) ? AtomicExpansionKind::CmpXChg
+                                 : AtomicExpansionKind::None;
 }
 
 TargetLowering::AtomicExpansionKind
@@ -27312,6 +27326,32 @@
     Results.push_back(EFLAGS.getValue(1));
     return;
   }
+  case ISD::ATOMIC_LOAD: {
+    assert(N->getValueType(0) == MVT::i64 && "Unexpected VT!");
+    bool NoImplicitFloatOps =
+        DAG.getMachineFunction().getFunction().hasFnAttribute(
+            Attribute::NoImplicitFloat);
+    if (!Subtarget.useSoftFloat() && !NoImplicitFloatOps &&
+        Subtarget.hasSSE2()) {
+      auto *Node = cast<AtomicSDNode>(N);
+      // Use a VZEXT_LOAD which will be selected as MOVQ. Then extract the lower
+      // 64-bits.
+      SDVTList Tys = DAG.getVTList(MVT::v2i64, MVT::Other);
+      SDValue Ops[] = { Node->getChain(), Node->getBasePtr() };
+      SDValue Ld = DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops,
+                                           MVT::i64, Node->getMemOperand());
+      SDValue Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64, Ld,
+                                DAG.getIntPtrConstant(0, dl));
+      Results.push_back(Res);
+      Results.push_back(Ld.getValue(1));
+      return;
+    }
+    // TODO: Use MOVLPS when SSE1 is available?
+    // TODO: Use FILD/FISTP when X87 is available?
+    // Delegate to generic TypeLegalization. Situations we can really handle
+    // should have already been dealt with by AtomicExpandPass.cpp.
+    break;
+  }
   case ISD::ATOMIC_SWAP:
   case ISD::ATOMIC_LOAD_ADD:
   case ISD::ATOMIC_LOAD_SUB:
@@ -27323,11 +27363,10 @@
   case ISD::ATOMIC_LOAD_MAX:
   case ISD::ATOMIC_LOAD_UMIN:
   case ISD::ATOMIC_LOAD_UMAX:
-  case ISD::ATOMIC_LOAD: {
     // Delegate to generic TypeLegalization. Situations we can really handle
     // should have already been dealt with by AtomicExpandPass.cpp.
     break;
-  }
+
   case ISD::BITCAST: {
     assert(Subtarget.hasSSE2() && "Requires at least SSE2!");
     EVT DstVT = N->getValueType(0);
Index: llvm/trunk/test/CodeGen/X86/atomic-load-store-wide.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/atomic-load-store-wide.ll
+++ llvm/trunk/test/CodeGen/X86/atomic-load-store-wide.ll
@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mcpu=corei7 -mtriple=i686-- -verify-machineinstrs | FileCheck %s
+; RUN: llc < %s -mcpu=corei7 -mtriple=i686-- -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=SSE42
+; RUN: llc < %s -mtriple=i686-- -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=NOSSE
 
 ; 64-bit load/store on x86-32
 ; FIXME: The generated code can be substantially improved.
@@ -34,7 +35,40 @@
 }
 
 define i64 @test2(i64* %ptr) {
-; CHECK-LABEL: test2:
+; SSE42-LABEL: test2:
+; SSE42:       # %bb.0:
+; SSE42-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; SSE42-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; SSE42-NEXT:    movd %xmm0, %eax
+; SSE42-NEXT:    pextrd $1, %xmm0, %edx
+; SSE42-NEXT:    retl
+;
+; NOSSE-LABEL: test2:
+; NOSSE:       # %bb.0:
+; NOSSE-NEXT:    pushl %ebx
+; NOSSE-NEXT:    .cfi_def_cfa_offset 8
+; NOSSE-NEXT:    pushl %esi
+; NOSSE-NEXT:    .cfi_def_cfa_offset 12
+; NOSSE-NEXT:    .cfi_offset %esi, -12
+; NOSSE-NEXT:    .cfi_offset %ebx, -8
+; NOSSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; NOSSE-NEXT:    xorl %eax, %eax
+; NOSSE-NEXT:    xorl %edx, %edx
+; NOSSE-NEXT:    xorl %ecx, %ecx
+; NOSSE-NEXT:    xorl %ebx, %ebx
+; NOSSE-NEXT:    lock cmpxchg8b (%esi)
+; NOSSE-NEXT:    popl %esi
+; NOSSE-NEXT:    .cfi_def_cfa_offset 8
+; NOSSE-NEXT:    popl %ebx
+; NOSSE-NEXT:    .cfi_def_cfa_offset 4
+; NOSSE-NEXT:    retl
+  %val = load atomic i64, i64* %ptr seq_cst, align 8
+  ret i64 %val
+}
+
+; Same as test2, but with noimplicitfloat.
+define i64 @test3(i64* %ptr) noimplicitfloat {
+; CHECK-LABEL: test3:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    pushl %ebx
 ; CHECK-NEXT:    .cfi_def_cfa_offset 8
Index: llvm/trunk/test/CodeGen/X86/atomic-non-integer.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/atomic-non-integer.ll
+++ llvm/trunk/test/CodeGen/X86/atomic-non-integer.ll
@@ -212,30 +212,13 @@
 define double @load_double(double* %fptr) {
 ; X86-SSE-LABEL: load_double:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    pushl %ebx
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE-NEXT:    pushl %esi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
 ; X86-SSE-NEXT:    subl $12, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 24
-; X86-SSE-NEXT:    .cfi_offset %esi, -12
-; X86-SSE-NEXT:    .cfi_offset %ebx, -8
-; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SSE-NEXT:    xorl %eax, %eax
-; X86-SSE-NEXT:    xorl %edx, %edx
-; X86-SSE-NEXT:    xorl %ecx, %ecx
-; X86-SSE-NEXT:    xorl %ebx, %ebx
-; X86-SSE-NEXT:    lock cmpxchg8b (%esi)
-; X86-SSE-NEXT:    movd %edx, %xmm0
-; X86-SSE-NEXT:    movd %eax, %xmm1
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; X86-SSE-NEXT:    movq %xmm1, (%esp)
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 16
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; X86-SSE-NEXT:    movlps %xmm0, (%esp)
 ; X86-SSE-NEXT:    fldl (%esp)
 ; X86-SSE-NEXT:    addl $12, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
-; X86-SSE-NEXT:    popl %esi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE-NEXT:    popl %ebx
 ; X86-SSE-NEXT:    .cfi_def_cfa_offset 4
 ; X86-SSE-NEXT:    retl
 ;
@@ -440,30 +423,13 @@
 define double @load_double_seq_cst(double* %fptr) {
 ; X86-SSE-LABEL: load_double_seq_cst:
 ; X86-SSE:       # %bb.0:
-; X86-SSE-NEXT:    pushl %ebx
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE-NEXT:    pushl %esi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
 ; X86-SSE-NEXT:    subl $12, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 24
-; X86-SSE-NEXT:    .cfi_offset %esi, -12
-; X86-SSE-NEXT:    .cfi_offset %ebx, -8
-; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X86-SSE-NEXT:    xorl %eax, %eax
-; X86-SSE-NEXT:    xorl %edx, %edx
-; X86-SSE-NEXT:    xorl %ecx, %ecx
-; X86-SSE-NEXT:    xorl %ebx, %ebx
-; X86-SSE-NEXT:    lock cmpxchg8b (%esi)
-; X86-SSE-NEXT:    movd %edx, %xmm0
-; X86-SSE-NEXT:    movd %eax, %xmm1
-; X86-SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
-; X86-SSE-NEXT:    movq %xmm1, (%esp)
+; X86-SSE-NEXT:    .cfi_def_cfa_offset 16
+; X86-SSE-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-SSE-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; X86-SSE-NEXT:    movlps %xmm0, (%esp)
 ; X86-SSE-NEXT:    fldl (%esp)
 ; X86-SSE-NEXT:    addl $12, %esp
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 12
-; X86-SSE-NEXT:    popl %esi
-; X86-SSE-NEXT:    .cfi_def_cfa_offset 8
-; X86-SSE-NEXT:    popl %ebx
 ; X86-SSE-NEXT:    .cfi_def_cfa_offset 4
 ; X86-SSE-NEXT:    retl
 ;