diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index f6b5dd30460..623fa510f72 100644
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -711,10 +711,6 @@ X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned i,
       { X86::CMP32rr,         X86::CMP32rm },
       { X86::CMP64rr,         X86::CMP64rm },
       { X86::CMP8rr,          X86::CMP8rm },
-      { X86::CMPPDrri,        X86::CMPPDrmi },
-      { X86::CMPPSrri,        X86::CMPPSrmi },
-      { X86::CMPSDrr,         X86::CMPSDrm },
-      { X86::CMPSSrr,         X86::CMPSSrm },
       { X86::CVTSD2SSrr,      X86::CVTSD2SSrm },
       { X86::CVTSI2SD64rr,    X86::CVTSI2SD64rm },
       { X86::CVTSI2SDrr,      X86::CVTSI2SDrm },
@@ -795,6 +791,20 @@ X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned i,
       { X86::PSHUFHWri,       X86::PSHUFHWmi },
       { X86::PSHUFLWri,       X86::PSHUFLWmi },
       { X86::PsMOVZX64rr32,   X86::PsMOVZX64rm32 },
+      { X86::RCPPSr,          X86::RCPPSm },
+      { X86::RCPPSr_Int,      X86::RCPPSm_Int },
+      { X86::RSQRTPSr,        X86::RSQRTPSm },
+      { X86::RSQRTPSr_Int,    X86::RSQRTPSm_Int },
+      { X86::RSQRTSSr,        X86::RSQRTSSm },
+      { X86::RSQRTSSr_Int,    X86::RSQRTSSm_Int },
+      { X86::SQRTPDr,         X86::SQRTPDm },
+      { X86::SQRTPDr_Int,     X86::SQRTPDm_Int },
+      { X86::SQRTPSr,         X86::SQRTPSm },
+      { X86::SQRTPSr_Int,     X86::SQRTPSm_Int },
+      { X86::SQRTSDr,         X86::SQRTSDm },
+      { X86::SQRTSDr_Int,     X86::SQRTSDm_Int },
+      { X86::SQRTSSr,         X86::SQRTSSm },
+      { X86::SQRTSSr_Int,     X86::SQRTSSm_Int },
       { X86::TEST16rr,        X86::TEST16rm },
       { X86::TEST32rr,        X86::TEST32rm },
       { X86::TEST64rr,        X86::TEST64rm },
@@ -875,6 +885,10 @@ X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned i,
       { X86::CMOVS16rr,       X86::CMOVS16rm },
       { X86::CMOVS32rr,       X86::CMOVS32rm },
       { X86::CMOVS64rr,       X86::CMOVS64rm },
+      { X86::CMPPDrri,        X86::CMPPDrmi },
+      { X86::CMPPSrri,        X86::CMPPSrmi },
+      { X86::CMPSDrr,         X86::CMPSDrm },
+      { X86::CMPSSrr,         X86::CMPSSrm },
       { X86::DIVPDrr,         X86::DIVPDrm },
       { X86::DIVPSrr,         X86::DIVPSrm },
       { X86::DIVSDrr,         X86::DIVSDrm },
@@ -965,24 +979,10 @@ X86RegisterInfo::foldMemoryOperand(MachineInstr *MI, unsigned i,
       { X86::PUNPCKLQDQrr,    X86::PUNPCKLQDQrm },
       { X86::PUNPCKLWDrr,     X86::PUNPCKLWDrm },
       { X86::PXORrr,          X86::PXORrm },
-      { X86::RCPPSr,          X86::RCPPSm },
-      { X86::RCPPSr_Int,      X86::RCPPSm_Int },
-      { X86::RSQRTPSr,        X86::RSQRTPSm },
-      { X86::RSQRTPSr_Int,    X86::RSQRTPSm_Int },
-      { X86::RSQRTSSr,        X86::RSQRTSSm },
-      { X86::RSQRTSSr_Int,    X86::RSQRTSSm_Int },
       { X86::SBB32rr,         X86::SBB32rm },
       { X86::SBB64rr,         X86::SBB64rm },
       { X86::SHUFPDrri,       X86::SHUFPDrmi },
       { X86::SHUFPSrri,       X86::SHUFPSrmi },
-      { X86::SQRTPDr,         X86::SQRTPDm },
-      { X86::SQRTPDr_Int,     X86::SQRTPDm_Int },
-      { X86::SQRTPSr,         X86::SQRTPSm },
-      { X86::SQRTPSr_Int,     X86::SQRTPSm_Int },
-      { X86::SQRTSDr,         X86::SQRTSDm },
-      { X86::SQRTSDr_Int,     X86::SQRTSDm_Int },
-      { X86::SQRTSSr,         X86::SQRTSSm },
-      { X86::SQRTSSr_Int,     X86::SQRTSSm_Int },
       { X86::SUB16rr,         X86::SUB16rm },
       { X86::SUB32rr,         X86::SUB32rm },
       { X86::SUB64rr,         X86::SUB64rm },
diff --git a/test/CodeGen/X86/sse_reload_fold.ll b/test/CodeGen/X86/sse_reload_fold.ll
new file mode 100644
index 00000000000..1100eb7c467
--- /dev/null
+++ b/test/CodeGen/X86/sse_reload_fold.ll
@@ -0,0 +1,124 @@
+; RUN: llvm-as < %s | \
+; RUN: llc -march=x86-64 -mattr=+64bit,+sse3 -print-failed-fuse-candidates |& \
+; RUN:   grep fail | count 1
+
+declare void @bar()
+declare <4 x float> @qux()
+declare <2 x double> @pin()
+declare float @llvm.sqrt.f32(float)
+declare double @llvm.sqrt.f64(double)
+
+declare <4 x float> @llvm.x86.sse.rsqrt.ps(<4 x float>)
+declare <4 x float> @llvm.x86.sse.sqrt.ps(<4 x float>)
+declare <4 x float> @llvm.x86.sse.rcp.ps(<4 x float>)
+declare <4 x float> @llvm.x86.sse.min.ps(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse.max.ps(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse.cmp.ps(<4 x float>, <4 x float>, i8)
+declare <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float>, <4 x float>)
+declare <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float>, <4 x float>)
+declare <2 x double> @llvm.x86.sse2.sqrt.pd(<2 x double>)
+declare <2 x double> @llvm.x86.sse2.min.pd(<2 x double>, <2 x double>)
+declare <2 x double> @llvm.x86.sse2.max.pd(<2 x double>, <2 x double>)
+declare <2 x double> @llvm.x86.sse2.cmp.pd(<2 x double>, <2 x double>, i8)
+declare <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double>, <2 x double>)
+declare <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double>, <2 x double>)
+declare <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double>, <2 x double>)
+
+define float @foo(float %f) {
+  call void @bar()
+  %t = call float @llvm.sqrt.f32(float %f)
+  ret float %t
+}
+define double @doo(double %f) {
+  call void @bar()
+  %t = call double @llvm.sqrt.f64(double %f)
+  ret double %t
+}
+define <4 x float> @a0(<4 x float> %f) {
+  call void @bar()
+  %t = call <4 x float> @llvm.x86.sse.rsqrt.ps(<4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @a1(<4 x float> %f) {
+  call void @bar()
+  %t = call <4 x float> @llvm.x86.sse.sqrt.ps(<4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @a2(<4 x float> %f) {
+  call void @bar()
+  %t = call <4 x float> @llvm.x86.sse.rcp.ps(<4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @b3(<4 x float> %f) {
+  %y = call <4 x float> @qux()
+  %t = call <4 x float> @llvm.x86.sse.min.ps(<4 x float> %y, <4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @b4(<4 x float> %f) {
+  %y = call <4 x float> @qux()
+  %t = call <4 x float> @llvm.x86.sse.max.ps(<4 x float> %y, <4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @b5(<4 x float> %f) {
+  %y = call <4 x float> @qux()
+  %t = call <4 x float> @llvm.x86.sse.cmp.ps(<4 x float> %y, <4 x float> %f, i8 7)
+  ret <4 x float> %t
+}
+define <4 x float> @b6(<4 x float> %f) {
+  %y = call <4 x float> @qux()
+  %t = call <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float> %y, <4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @b7(<4 x float> %f) {
+  %y = call <4 x float> @qux()
+  %t = call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %y, <4 x float> %f)
+  ret <4 x float> %t
+}
+define <4 x float> @b8(<4 x float> %f) {
+  %y = call <4 x float> @qux()
+  %t = call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %y, <4 x float> %f)
+  ret <4 x float> %t
+}
+define <2 x double> @c1(<2 x double> %f) {
+  call void @bar()
+  %t = call <2 x double> @llvm.x86.sse2.sqrt.pd(<2 x double> %f)
+  ret <2 x double> %t
+}
+define <2 x double> @d3(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse2.min.pd(<2 x double> %y, <2 x double> %f)
+  ret <2 x double> %t
+}
+define <2 x double> @d4(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse2.max.pd(<2 x double> %y, <2 x double> %f)
+  ret <2 x double> %t
+}
+define <2 x double> @d5(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse2.cmp.pd(<2 x double> %y, <2 x double> %f, i8 7)
+  ret <2 x double> %t
+}
+define <2 x double> @d6(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double> %y, <2 x double> %f)
+  ret <2 x double> %t
+}
+define <2 x double> @d7(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %y, <2 x double> %f)
+  ret <2 x double> %t
+}
+define <2 x double> @d8(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %y, <2 x double> %f)
+  ret <2 x double> %t
+}
+
+; This one should fail to fuse.
+define <2 x double> @z0(<2 x double> %f) {
+  %y = call <2 x double> @pin()
+  %t = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %f, <2 x double> %y)
+  ret <2 x double> %t
+}