Index: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
@@ -34306,8 +34306,11 @@
          ((Subtarget.hasAVX() && BitWidth >= 32) || Subtarget.hasAVX2()))))
     return SDValue();
 
-  // Don't bother performing this for 2-element vectors.
-  if (Match.getValueType().getVectorNumElements() <= 2)
+  // Make sure this isn't a vector of 1 element. The perf win from using MOVMSK
+  // diminishes with less elements in the reduction, but it is generally better
+  // to get the comparison over to the GPRs as soon as possible to reduce the
+  // number of vector ops.
+  if (Match.getValueType().getVectorNumElements() < 2)
     return SDValue();
 
   // Check that we are extracting a reduction of all sign bits.
Index: llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll
+++ llvm/trunk/test/CodeGen/X86/vector-compare-all_of.ll
@@ -8,17 +8,21 @@
 ; SSE-LABEL: test_v2f64_sext:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    movq %xmm0, %rax
+; SSE-NEXT:    movmskpd %xmm1, %ecx
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:    cmpl $3, %ecx
+; SSE-NEXT:    sete %al
+; SSE-NEXT:    negq %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2f64_sext:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vandpd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovq %xmm0, %rax
+; AVX-NEXT:    vmovmskpd %xmm0, %ecx
+; AVX-NEXT:    xorl %eax, %eax
+; AVX-NEXT:    cmpl $3, %ecx
+; AVX-NEXT:    sete %al
+; AVX-NEXT:    negq %rax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2f64_sext:
@@ -42,9 +46,11 @@
 ; SSE-NEXT:    cmpltpd %xmm1, %xmm3
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm2
 ; SSE-NEXT:    andpd %xmm3, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE-NEXT:    pand %xmm2, %xmm0
-; SSE-NEXT:    movq %xmm0, %rax
+; SSE-NEXT:    movmskpd %xmm2, %ecx
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:    cmpl $3, %ecx
+; SSE-NEXT:    sete %al
+; SSE-NEXT:    negq %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4f64_sext:
@@ -273,17 +279,21 @@
 ; SSE-LABEL: test_v2i64_sext:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtq %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movq %xmm1, %rax
+; SSE-NEXT:    movmskpd %xmm0, %ecx
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:    cmpl $3, %ecx
+; SSE-NEXT:    sete %al
+; SSE-NEXT:    negq %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i64_sext:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovq %xmm0, %rax
+; AVX-NEXT:    vmovmskpd %xmm0, %ecx
+; AVX-NEXT:    xorl %eax, %eax
+; AVX-NEXT:    cmpl $3, %ecx
+; AVX-NEXT:    sete %al
+; AVX-NEXT:    negq %rax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i64_sext:
@@ -307,9 +317,11 @@
 ; SSE-NEXT:    pcmpgtq %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtq %xmm2, %xmm0
 ; SSE-NEXT:    pand %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    pand %xmm0, %xmm1
-; SSE-NEXT:    movq %xmm1, %rax
+; SSE-NEXT:    movmskpd %xmm0, %ecx
+; SSE-NEXT:    xorl %eax, %eax
+; SSE-NEXT:    cmpl $3, %ecx
+; SSE-NEXT:    sete %al
+; SSE-NEXT:    negq %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v4i64_sext:
Index: llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll
+++ llvm/trunk/test/CodeGen/X86/vector-compare-any_of.ll
@@ -8,17 +8,17 @@
 ; SSE-LABEL: test_v2f64_sext:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
-; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    movq %xmm0, %rax
+; SSE-NEXT:    movmskpd %xmm1, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbq %rax, %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2f64_sext:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm0[1,0]
-; AVX-NEXT:    vorpd %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovq %xmm0, %rax
+; AVX-NEXT:    vmovmskpd %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbq %rax, %rax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2f64_sext:
@@ -42,9 +42,9 @@
 ; SSE-NEXT:    cmpltpd %xmm1, %xmm3
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm2
 ; SSE-NEXT:    orpd %xmm3, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[2,3,0,1]
-; SSE-NEXT:    por %xmm2, %xmm0
-; SSE-NEXT:    movq %xmm0, %rax
+; SSE-NEXT:    movmskpd %xmm2, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbq %rax, %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v4f64_sext:
@@ -255,17 +255,17 @@
 ; SSE-LABEL: test_v2i64_sext:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    pcmpgtq %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movq %xmm1, %rax
+; SSE-NEXT:    movmskpd %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbq %rax, %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_v2i64_sext:
 ; AVX:       # %bb.0:
 ; AVX-NEXT:    vpcmpgtq %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
-; AVX-NEXT:    vmovq %xmm0, %rax
+; AVX-NEXT:    vmovmskpd %xmm0, %eax
+; AVX-NEXT:    negl %eax
+; AVX-NEXT:    sbbq %rax, %rax
 ; AVX-NEXT:    retq
 ;
 ; AVX512-LABEL: test_v2i64_sext:
@@ -289,9 +289,9 @@
 ; SSE-NEXT:    pcmpgtq %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtq %xmm2, %xmm0
 ; SSE-NEXT:    por %xmm1, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE-NEXT:    por %xmm0, %xmm1
-; SSE-NEXT:    movq %xmm1, %rax
+; SSE-NEXT:    movmskpd %xmm0, %eax
+; SSE-NEXT:    negl %eax
+; SSE-NEXT:    sbbq %rax, %rax
 ; SSE-NEXT:    retq
 ;
 ; AVX1-LABEL: test_v4i64_sext: