Index: lib/CodeGen/PeepholeOptimizer.cpp
===================================================================
--- lib/CodeGen/PeepholeOptimizer.cpp
+++ lib/CodeGen/PeepholeOptimizer.cpp
@@ -1236,14 +1236,13 @@
 
       // If there exists an instruction which belongs to the following
       // categories, we will discard the load candidates.
+      if (MI->mayStore() || MI->isCall() || MI->hasUnmodeledSideEffects())
+        FoldAsLoadDefCandidates.clear();
+
       if (MI->isPosition() || MI->isPHI() || MI->isImplicitDef() ||
           MI->isKill() || MI->isInlineAsm() ||
-          MI->hasUnmodeledSideEffects()) {
-        FoldAsLoadDefCandidates.clear();
+          MI->hasUnmodeledSideEffects())
         continue;
-      }
-      if (MI->mayStore() || MI->isCall())
-        FoldAsLoadDefCandidates.clear();
 
       if ((isUncoalescableCopy(*MI) &&
            optimizeUncoalescableCopy(MI, LocalMIs)) ||
Index: test/CodeGen/X86/avx-cvt.ll
===================================================================
--- test/CodeGen/X86/avx-cvt.ll
+++ test/CodeGen/X86/avx-cvt.ll
@@ -113,8 +113,7 @@
 define void @fpext() nounwind uwtable {
 ; CHECK-LABEL: fpext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
-; CHECK-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
+; CHECK-NEXT:    vcvtss2sd -{{[0-9]+}}(%rsp), %xmm0, %xmm0
 ; CHECK-NEXT:    vmovsd %xmm0, -{{[0-9]+}}(%rsp)
 ; CHECK-NEXT:    retq
   %f = alloca float, align 4
Index: test/CodeGen/X86/shift-bmi2.ll
===================================================================
--- test/CodeGen/X86/shift-bmi2.ll
+++ test/CodeGen/X86/shift-bmi2.ll
@@ -30,11 +30,10 @@
   %x = load i32, i32* %p
   %shl = shl i32 %x, %shamt
 ; BMI2: shl32p
-; Source order scheduling prevents folding, rdar:14208996.
-; BMI2: shlxl %{{.+}}, %{{.+}}, %{{.+}}
+; BMI2: shlxl %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI2: ret
 ; BMI264: shl32p
-; BMI264: shlxl %{{.+}}, %{{.+}}, %{{.+}}
+; BMI264: shlxl %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI264: ret
   ret i32 %shl
 }
@@ -75,7 +74,7 @@
   %x = load i64, i64* %p
   %shl = shl i64 %x, %shamt
 ; BMI264: shl64p
-; BMI264: shlxq %{{.+}}, %{{.+}}, %{{.+}}
+; BMI264: shlxq %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI264: ret
   ret i64 %shl
 }
@@ -107,11 +106,10 @@
   %x = load i32, i32* %p
   %shl = lshr i32 %x, %shamt
 ; BMI2: lshr32p
-; Source order scheduling prevents folding, rdar:14208996.
-; BMI2: shrxl %{{.+}}, %{{.+}}, %{{.+}}
+; BMI2: shrxl %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI2: ret
 ; BMI264: lshr32p
-; BMI264: shrxl %{{.+}}, %{{.+}}, %{{.+}}
+; BMI264: shrxl %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI264: ret
   ret i32 %shl
 }
@@ -130,7 +128,7 @@
   %x = load i64, i64* %p
   %shl = lshr i64 %x, %shamt
 ; BMI264: lshr64p
-; BMI264: shrxq %{{.+}}, %{{.+}}, %{{.+}}
+; BMI264: shrxq %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI264: ret
   ret i64 %shl
 }
@@ -153,10 +151,10 @@
   %shl = ashr i32 %x, %shamt
 ; BMI2: ashr32p
 ; Source order scheduling prevents folding, rdar:14208996.
-; BMI2: sarxl %{{.+}}, %{{.+}}, %{{.+}}
+; BMI2: sarxl %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI2: ret
 ; BMI264: ashr32p
-; BMI264: sarxl %{{.+}}, %{{.+}}, %{{.+}}
+; BMI264: sarxl %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI264: ret
   ret i32 %shl
 }
@@ -175,7 +173,7 @@
   %x = load i64, i64* %p
   %shl = ashr i64 %x, %shamt
 ; BMI264: ashr64p
-; BMI264: sarxq %{{.+}}, %{{.+}}, %{{.+}}
+; BMI264: sarxq %{{.+}}, ({{.+}}), %{{.+}}
 ; BMI264: ret
   ret i64 %shl
 }