Index: llvm/trunk/lib/Transforms/Scalar/LICM.cpp
===================================================================
--- llvm/trunk/lib/Transforms/Scalar/LICM.cpp
+++ llvm/trunk/lib/Transforms/Scalar/LICM.cpp
@@ -1924,9 +1924,21 @@
         SawUnorderedAtomic |= Load->isAtomic();
         SawNotAtomic |= !Load->isAtomic();
 
-        if (!DereferenceableInPH)
-          DereferenceableInPH = isSafeToExecuteUnconditionally(
-              *Load, DT, CurLoop, SafetyInfo, ORE, Preheader->getTerminator());
+        unsigned InstAlignment = Load->getAlignment();
+        if (!InstAlignment)
+          InstAlignment =
+              MDL.getABITypeAlignment(Load->getType());
+
+        // Note that proving a load safe to speculate requires proving
+        // sufficient alignment at the target location.  Proving it guaranteed
+        // to execute does as well.  Thus we can increase our guaranteed
+        // alignment as well. 
+        if (!DereferenceableInPH || (InstAlignment > Alignment))
+          if (isSafeToExecuteUnconditionally(*Load, DT, CurLoop, SafetyInfo,
+                                             ORE, Preheader->getTerminator())) {
+            DereferenceableInPH = true;
+            Alignment = std::max(Alignment, InstAlignment);
+          }
       } else if (const StoreInst *Store = dyn_cast<StoreInst>(UI)) {
         // Stores *of* the pointer are not interesting, only stores *to* the
         // pointer.
@@ -1997,6 +2009,14 @@
   if (SawUnorderedAtomic && SawNotAtomic)
     return false;
 
+  // If we're inserting an atomic load in the preheader, we must be able to
+  // lower it.  We're only guaranteed to be able to lower naturally aligned
+  // atomics.
+  auto *SomePtrElemType = SomePtr->getType()->getPointerElementType();
+  if (SawUnorderedAtomic &&
+      Alignment < MDL.getTypeStoreSize(SomePtrElemType))
+    return false;
+
   // If we couldn't prove we can hoist the load, bail.
   if (!DereferenceableInPH)
     return false;
Index: llvm/trunk/test/Transforms/LICM/promote-tls.ll
===================================================================
--- llvm/trunk/test/Transforms/LICM/promote-tls.ll
+++ llvm/trunk/test/Transforms/LICM/promote-tls.ll
@@ -22,7 +22,7 @@
 
 for.body.lr.ph:                                   ; preds = %entry
 ; CHECK-LABEL: for.body.lr.ph:
-; CHECK-NEXT: %addr.promoted = load i32, i32* %addr, align 1
+; CHECK-NEXT: %addr.promoted = load i32, i32* %addr, align 4
   br label %for.header
 
 for.header:
@@ -35,7 +35,7 @@
 
 early-exit:
 ; CHECK-LABEL: early-exit:
-; CHECK: store i32 %new1.lcssa, i32* %addr, align 1
+; CHECK: store i32 %new1.lcssa, i32* %addr, align 4
   ret i32* null
 
 for.body:
@@ -47,7 +47,7 @@
 
 for.cond.for.end_crit_edge:                       ; preds = %for.body
 ; CHECK-LABEL: for.cond.for.end_crit_edge:
-; CHECK: store i32 %new.lcssa, i32* %addr, align 1
+; CHECK: store i32 %new.lcssa, i32* %addr, align 4
   %split = phi i32* [ %addr, %for.body ]
   ret i32* null
 }
@@ -62,7 +62,7 @@
 
 for.body.lr.ph:                                   ; preds = %entry
 ; CHECK-LABEL: for.body.lr.ph:
-; CHECK-NEXT: %addr.promoted = load i32, i32* %addr, align 1
+; CHECK-NEXT: %addr.promoted = load i32, i32* %addr, align 4
   br label %for.header
 
 for.header:
@@ -75,7 +75,7 @@
 
 early-exit:
 ; CHECK-LABEL: early-exit:
-; CHECK: store i32 %new1.lcssa, i32* %addr, align 1
+; CHECK: store i32 %new1.lcssa, i32* %addr, align 4
   ret i32* null
 
 for.body:
@@ -87,7 +87,7 @@
 
 for.cond.for.end_crit_edge:                       ; preds = %for.body
 ; CHECK-LABEL: for.cond.for.end_crit_edge:
-; CHECK: store i32 %new.lcssa, i32* %addr, align 1
+; CHECK: store i32 %new.lcssa, i32* %addr, align 4
   %split = phi i32* [ %addr, %for.body ]
   ret i32* null
 }
Index: llvm/trunk/test/Transforms/LICM/scalar-promote-unwind.ll
===================================================================
--- llvm/trunk/test/Transforms/LICM/scalar-promote-unwind.ll
+++ llvm/trunk/test/Transforms/LICM/scalar-promote-unwind.ll
@@ -74,7 +74,7 @@
 entry:
 ; CHECK-LABEL: entry:
 ; CHECK-NEXT:  %a = alloca i32
-; CHECK-NEXT:  %a.promoted = load i32, i32* %a, align 1
+; CHECK-NEXT:  %a.promoted = load i32, i32* %a, align 4
   %a = alloca i32
   br label %for.body
 
@@ -90,20 +90,18 @@
 
 for.cond.cleanup:
 ; CHECK-LABEL: for.cond.cleanup:
-; CHECK: store i32 %add.lcssa, i32* %a, align 1
+; CHECK: store i32 %add.lcssa, i32* %a, align 4
 ; CHECK-NEXT: ret void
   ret void
 }
 
 ;; Same as test3, but with unordered atomics
-;; FIXME: doing the transform w/o alignment here is wrong since we're
-;; creating an unaligned atomic which we may not be able to lower.
 define void @test3b(i1 zeroext %y) uwtable {
 ; CHECK-LABEL: @test3
 entry:
 ; CHECK-LABEL: entry:
 ; CHECK-NEXT:  %a = alloca i32
-; CHECK-NEXT:  %a.promoted = load atomic i32, i32* %a unordered, align 1
+; CHECK-NEXT:  %a.promoted = load atomic i32, i32* %a unordered, align 4
   %a = alloca i32
   br label %for.body
 
@@ -119,7 +117,7 @@
 
 for.cond.cleanup:
 ; CHECK-LABEL: for.cond.cleanup:
-; CHECK: store atomic i32 %add.lcssa, i32* %a unordered, align 1
+; CHECK: store atomic i32 %add.lcssa, i32* %a unordered, align 4
 ; CHECK-NEXT: ret void
   ret void
 }