1 files changed, 103 insertions, 60 deletions
diff --git a/llvm/test/CodeGen/X86/bitcast-and-setcc-512.ll b/llvm/test/CodeGen/X86/bitcast-and-setcc-512.ll
index a63fe5ab466..183e32a518b 100644
--- a/llvm/test/CodeGen/X86/bitcast-and-setcc-512.ll
+++ b/llvm/test/CodeGen/X86/bitcast-and-setcc-512.ll
@@ -13,22 +13,38 @@ define i8 @v8i64(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm10
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    pcmpgtq %xmm7, %xmm3
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm3[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,1,0,2,4,5,6,7]
 ; SSE-NEXT:    pcmpgtq %xmm6, %xmm2
-; SSE-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm3[0,2]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,0,2,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
 ; SSE-NEXT:    pcmpgtq %xmm5, %xmm1
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    pcmpgtq %xmm4, %xmm0
-; SSE-NEXT:    shufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2]
-; SSE-NEXT:    packssdw %xmm2, %xmm0
+; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
 ; SSE-NEXT:    pcmpgtq {{[0-9]+}}(%rsp), %xmm11
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm11[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,0,2,4,5,6,7]
 ; SSE-NEXT:    pcmpgtq {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,2],xmm11[0,2]
+; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm10[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[0,1,0,2,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
 ; SSE-NEXT:    pcmpgtq {{[0-9]+}}(%rsp), %xmm9
+; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm9[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    pcmpgtq {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[0,2],xmm9[0,2]
-; SSE-NEXT:    packssdw %xmm10, %xmm8
-; SSE-NEXT:    pand %xmm0, %xmm8
-; SSE-NEXT:    packsswb %xmm0, %xmm8
-; SSE-NEXT:    pmovmskb %xmm8, %eax
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm8[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm3[0,2,2,3,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    packsswb %xmm0, %xmm3
+; SSE-NEXT:    pmovmskb %xmm3, %eax
 ; SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE-NEXT:    retq
 ;
@@ -38,30 +54,27 @@ define i8 @v8i64(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm9
 ; AVX1-NEXT:    vpcmpgtq %xmm8, %xmm9, %xmm8
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm1
-; AVX1-NEXT:    vpackssdw %xmm8, %xmm1, %xmm1
-; AVX1-NEXT:    vmovdqa {{.*#+}} xmm8 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX1-NEXT:    vpshufb %xmm8, %xmm1, %xmm9
+; AVX1-NEXT:    vpackssdw %xmm8, %xmm1, %xmm8
 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm1, %xmm1
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb %xmm8, %xmm0, %xmm0
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm9[0]
+; AVX1-NEXT:    vpackssdw %xmm8, %xmm0, %xmm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm7, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm5, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpcmpgtq %xmm7, %xmm5, %xmm2
 ; AVX1-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
-; AVX1-NEXT:    vpshufb %xmm8, %xmm1, %xmm1
 ; AVX1-NEXT:    vextractf128 $1, %ymm6, %xmm2
 ; AVX1-NEXT:    vextractf128 $1, %ymm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm3, %xmm2
 ; AVX1-NEXT:    vpcmpgtq %xmm6, %xmm4, %xmm3
 ; AVX1-NEXT:    vpackssdw %xmm2, %xmm3, %xmm2
-; AVX1-NEXT:    vpshufb %xmm8, %xmm2, %xmm2
-; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX1-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
 ; AVX1-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpsllw $15, %xmm0, %xmm0
+; AVX1-NEXT:    vpsraw $15, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    # kill: def $al killed $al killed $eax
@@ -73,23 +86,20 @@ define i8 @v8i64(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
 ; AVX2-NEXT:    vpcmpgtq %ymm3, %ymm1, %ymm1
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
 ; AVX2-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
 ; AVX2-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpgtq %ymm7, %ymm5, %ymm1
 ; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm2
 ; AVX2-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX2-NEXT:    vpcmpgtq %ymm6, %ymm4, %ymm2
-; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm4
-; AVX2-NEXT:    vpackssdw %xmm4, %xmm2, %xmm2
-; AVX2-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
+; AVX2-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; AVX2-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpsllw $15, %xmm0, %xmm0
+; AVX2-NEXT:    vpsraw $15, %xmm0, %xmm0
 ; AVX2-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX2-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
@@ -128,22 +138,38 @@ define i8 @v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x double>
 ; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm10
 ; SSE-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    cmpltpd %xmm3, %xmm7
+; SSE-NEXT:    shufps {{.*#+}} xmm7 = xmm7[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm7[0,1,0,2,4,5,6,7]
 ; SSE-NEXT:    cmpltpd %xmm2, %xmm6
-; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2],xmm7[0,2]
+; SSE-NEXT:    shufps {{.*#+}} xmm6 = xmm6[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm6[0,1,0,2,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
 ; SSE-NEXT:    cmpltpd %xmm1, %xmm5
+; SSE-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm5[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    cmpltpd %xmm0, %xmm4
-; SSE-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2],xmm5[0,2]
-; SSE-NEXT:    packssdw %xmm6, %xmm4
+; SSE-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm0 = xmm4[0,2,2,3,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm11
+; SSE-NEXT:    shufps {{.*#+}} xmm11 = xmm11[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm11[0,1,0,2,4,5,6,7]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,2],xmm11[0,2]
+; SSE-NEXT:    shufps {{.*#+}} xmm10 = xmm10[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm2 = xmm10[0,1,0,2,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm9
+; SSE-NEXT:    shufps {{.*#+}} xmm9 = xmm9[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm1 = xmm9[0,2,2,3,4,5,6,7]
 ; SSE-NEXT:    cmpltpd {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[0,2],xmm9[0,2]
-; SSE-NEXT:    packssdw %xmm10, %xmm8
-; SSE-NEXT:    pand %xmm4, %xmm8
-; SSE-NEXT:    packsswb %xmm0, %xmm8
-; SSE-NEXT:    pmovmskb %xmm8, %eax
+; SSE-NEXT:    shufps {{.*#+}} xmm8 = xmm8[0,2,2,3]
+; SSE-NEXT:    pshuflw {{.*#+}} xmm3 = xmm8[0,2,2,3,4,5,6,7]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm1[0],xmm3[1],xmm1[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm3 = xmm3[0,1,2,3],xmm2[4,5,6,7]
+; SSE-NEXT:    pand %xmm0, %xmm3
+; SSE-NEXT:    packsswb %xmm0, %xmm3
+; SSE-NEXT:    pmovmskb %xmm3, %eax
 ; SSE-NEXT:    # kill: def $al killed $al killed $eax
 ; SSE-NEXT:    retq
 ;
@@ -152,23 +178,20 @@ define i8 @v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x double>
 ; AVX12-NEXT:    vcmpltpd %ymm1, %ymm3, %ymm1
 ; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm3
 ; AVX12-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; AVX12-NEXT:    vmovdqa {{.*#+}} xmm3 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
-; AVX12-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX12-NEXT:    vcmpltpd %ymm0, %ymm2, %ymm0
 ; AVX12-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX12-NEXT:    vpackssdw %xmm2, %xmm0, %xmm0
-; AVX12-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
-; AVX12-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; AVX12-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
 ; AVX12-NEXT:    vcmpltpd %ymm5, %ymm7, %ymm1
 ; AVX12-NEXT:    vextractf128 $1, %ymm1, %xmm2
 ; AVX12-NEXT:    vpackssdw %xmm2, %xmm1, %xmm1
-; AVX12-NEXT:    vpshufb %xmm3, %xmm1, %xmm1
 ; AVX12-NEXT:    vcmpltpd %ymm4, %ymm6, %ymm2
-; AVX12-NEXT:    vextractf128 $1, %ymm2, %xmm4
-; AVX12-NEXT:    vpackssdw %xmm4, %xmm2, %xmm2
-; AVX12-NEXT:    vpshufb %xmm3, %xmm2, %xmm2
-; AVX12-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
+; AVX12-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; AVX12-NEXT:    vpackssdw %xmm3, %xmm2, %xmm2
+; AVX12-NEXT:    vpackssdw %xmm1, %xmm2, %xmm1
 ; AVX12-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX12-NEXT:    vpsllw $15, %xmm0, %xmm0
+; AVX12-NEXT:    vpsraw $15, %xmm0, %xmm0
 ; AVX12-NEXT:    vpacksswb %xmm0, %xmm0, %xmm0
 ; AVX12-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX12-NEXT:    # kill: def $al killed $al killed $eax
@@ -317,23 +340,33 @@ define i16 @v16i32(<16 x i32> %a, <16 x i32> %b, <16 x i32> %c, <16 x i32> %d) {
 ; SSE-LABEL: v16i32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm10
+; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    movdqa {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    pcmpgtd %xmm7, %xmm3
+; SSE-NEXT:    movdqa {{.*#+}} xmm7 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
+; SSE-NEXT:    pshufb %xmm7, %xmm3
 ; SSE-NEXT:    pcmpgtd %xmm6, %xmm2
-; SSE-NEXT:    packssdw %xmm3, %xmm2
+; SSE-NEXT:    pshufb %xmm7, %xmm2
+; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
 ; SSE-NEXT:    pcmpgtd %xmm5, %xmm1
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
+; SSE-NEXT:    pshufb %xmm3, %xmm1
 ; SSE-NEXT:    pcmpgtd %xmm4, %xmm0
-; SSE-NEXT:    packssdw %xmm1, %xmm0
-; SSE-NEXT:    packsswb %xmm2, %xmm0
+; SSE-NEXT:    pshufb %xmm3, %xmm0
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7]
 ; SSE-NEXT:    pcmpgtd {{[0-9]+}}(%rsp), %xmm11
-; SSE-NEXT:    pcmpgtd {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    packssdw %xmm11, %xmm10
+; SSE-NEXT:    pshufb %xmm7, %xmm11
 ; SSE-NEXT:    pcmpgtd {{[0-9]+}}(%rsp), %xmm9
+; SSE-NEXT:    pshufb %xmm7, %xmm9
+; SSE-NEXT:    punpckldq {{.*#+}} xmm9 = xmm9[0],xmm11[0],xmm9[1],xmm11[1]
+; SSE-NEXT:    pcmpgtd {{[0-9]+}}(%rsp), %xmm10
+; SSE-NEXT:    pshufb %xmm3, %xmm10
 ; SSE-NEXT:    pcmpgtd {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    packssdw %xmm9, %xmm8
-; SSE-NEXT:    packsswb %xmm10, %xmm8
+; SSE-NEXT:    pshufb %xmm3, %xmm8
+; SSE-NEXT:    punpckldq {{.*#+}} xmm8 = xmm8[0],xmm10[0],xmm8[1],xmm10[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm9[4,5,6,7]
 ; SSE-NEXT:    pand %xmm0, %xmm8
 ; SSE-NEXT:    pmovmskb %xmm8, %eax
 ; SSE-NEXT:    # kill: def $ax killed $ax killed $eax
@@ -419,23 +452,33 @@ define i16 @v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x floa
 ; SSE-LABEL: v16f32:
 ; SSE:       # %bb.0:
 ; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm10
+; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm9
 ; SSE-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm11
 ; SSE-NEXT:    cmpltps %xmm3, %xmm7
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = <u,u,u,u,0,4,8,12,u,u,u,u,u,u,u,u>
+; SSE-NEXT:    pshufb %xmm3, %xmm7
 ; SSE-NEXT:    cmpltps %xmm2, %xmm6
-; SSE-NEXT:    packssdw %xmm7, %xmm6
+; SSE-NEXT:    pshufb %xmm3, %xmm6
+; SSE-NEXT:    punpckldq {{.*#+}} xmm6 = xmm6[0],xmm7[0],xmm6[1],xmm7[1]
 ; SSE-NEXT:    cmpltps %xmm1, %xmm5
+; SSE-NEXT:    movdqa {{.*#+}} xmm1 = <0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u>
+; SSE-NEXT:    pshufb %xmm1, %xmm5
 ; SSE-NEXT:    cmpltps %xmm0, %xmm4
-; SSE-NEXT:    packssdw %xmm5, %xmm4
-; SSE-NEXT:    packsswb %xmm6, %xmm4
+; SSE-NEXT:    pshufb %xmm1, %xmm4
+; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm5[0],xmm4[1],xmm5[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm6[4,5,6,7]
 ; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm11
-; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm10
-; SSE-NEXT:    packssdw %xmm11, %xmm10
+; SSE-NEXT:    pshufb %xmm3, %xmm11
 ; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm9
+; SSE-NEXT:    pshufb %xmm3, %xmm9
+; SSE-NEXT:    punpckldq {{.*#+}} xmm9 = xmm9[0],xmm11[0],xmm9[1],xmm11[1]
+; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm10
+; SSE-NEXT:    pshufb %xmm1, %xmm10
 ; SSE-NEXT:    cmpltps {{[0-9]+}}(%rsp), %xmm8
-; SSE-NEXT:    packssdw %xmm9, %xmm8
-; SSE-NEXT:    packsswb %xmm10, %xmm8
+; SSE-NEXT:    pshufb %xmm1, %xmm8
+; SSE-NEXT:    punpckldq {{.*#+}} xmm8 = xmm8[0],xmm10[0],xmm8[1],xmm10[1]
+; SSE-NEXT:    pblendw {{.*#+}} xmm8 = xmm8[0,1,2,3],xmm9[4,5,6,7]
 ; SSE-NEXT:    pand %xmm4, %xmm8
 ; SSE-NEXT:    pmovmskb %xmm8, %eax
 ; SSE-NEXT:    # kill: def $ax killed $ax killed $eax