summaryrefslogtreecommitdiff
path: root/test/CodeGen/X86/vector-shuffle-v1.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/CodeGen/X86/vector-shuffle-v1.ll')
-rw-r--r--test/CodeGen/X86/vector-shuffle-v1.ll185
1 files changed, 92 insertions, 93 deletions
diff --git a/test/CodeGen/X86/vector-shuffle-v1.ll b/test/CodeGen/X86/vector-shuffle-v1.ll
index 9c92ca756ebd..f3433ce834cd 100644
--- a/test/CodeGen/X86/vector-shuffle-v1.ll
+++ b/test/CodeGen/X86/vector-shuffle-v1.ll
@@ -48,9 +48,8 @@ define <2 x i1> @shuf2i1_1_2(<2 x i1> %a) {
; AVX512VL-NEXT: vptestmq %xmm0, %xmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %xmm0, %xmm0, %xmm0
; AVX512VL-NEXT: vmovdqa64 %xmm0, %xmm1 {%k1} {z}
-; AVX512VL-NEXT: movb $1, %al
-; AVX512VL-NEXT: kmovw %eax, %k1
-; AVX512VL-NEXT: vmovdqa64 %xmm0, %xmm2 {%k1} {z}
+; AVX512VL-NEXT: movq $-1, %rax
+; AVX512VL-NEXT: vmovq %rax, %xmm2
; AVX512VL-NEXT: vpalignr {{.*#+}} xmm1 = xmm1[8,9,10,11,12,13,14,15],xmm2[0,1,2,3,4,5,6,7]
; AVX512VL-NEXT: vpsllq $63, %xmm1, %xmm1
; AVX512VL-NEXT: vptestmq %xmm1, %xmm1, %k1
@@ -61,9 +60,8 @@ define <2 x i1> @shuf2i1_1_2(<2 x i1> %a) {
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: vpsllq $63, %xmm0, %xmm0
; VL_BW_DQ-NEXT: vptestmq %xmm0, %xmm0, %k0
-; VL_BW_DQ-NEXT: movb $1, %al
-; VL_BW_DQ-NEXT: kmovd %eax, %k1
-; VL_BW_DQ-NEXT: vpmovm2q %k1, %xmm0
+; VL_BW_DQ-NEXT: movq $-1, %rax
+; VL_BW_DQ-NEXT: vmovq %rax, %xmm0
; VL_BW_DQ-NEXT: vpmovm2q %k0, %xmm1
; VL_BW_DQ-NEXT: vpalignr {{.*#+}} xmm0 = xmm1[8,9,10,11,12,13,14,15],xmm0[0,1,2,3,4,5,6,7]
; VL_BW_DQ-NEXT: vpmovq2m %xmm0, %k0
@@ -123,12 +121,12 @@ define <8 x i1> @shuf8i1_3_6_1_0_3_7_7_0(<8 x i64> %a, <8 x i64> %b, <8 x i64> %
; AVX512VL-LABEL: shuf8i1_3_6_1_0_3_7_7_0:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpcmpeqq %zmm2, %zmm0, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm1 = [3,6,1,0,3,7,7,0]
-; AVX512VL-NEXT: vpermq %zmm0, %zmm1, %zmm0
-; AVX512VL-NEXT: vpsllq $63, %zmm0, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [3,6,1,0,3,7,7,0]
+; AVX512VL-NEXT: vpermd %ymm1, %ymm2, %ymm1
+; AVX512VL-NEXT: vpslld $31, %ymm1, %ymm1
+; AVX512VL-NEXT: vptestmd %ymm1, %ymm1, %k1
; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; AVX512VL-NEXT: vpmovdw %ymm0, %xmm0
; AVX512VL-NEXT: vzeroupper
@@ -137,10 +135,10 @@ define <8 x i1> @shuf8i1_3_6_1_0_3_7_7_0(<8 x i64> %a, <8 x i64> %b, <8 x i64> %
; VL_BW_DQ-LABEL: shuf8i1_3_6_1_0_3_7_7_0:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: vpcmpeqq %zmm2, %zmm0, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm1 = [3,6,1,0,3,7,7,0]
-; VL_BW_DQ-NEXT: vpermq %zmm0, %zmm1, %zmm0
-; VL_BW_DQ-NEXT: vpmovq2m %zmm0, %k0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vmovdqa {{.*#+}} ymm1 = [3,6,1,0,3,7,7,0]
+; VL_BW_DQ-NEXT: vpermd %ymm0, %ymm1, %ymm0
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: vpmovm2w %k0, %xmm0
; VL_BW_DQ-NEXT: vzeroupper
; VL_BW_DQ-NEXT: retq
@@ -250,12 +248,12 @@ define <8 x i1> @shuf8i1_u_2_u_u_2_u_2_u(i8 %a) {
; AVX512VL-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: kmovw %edi, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512VL-NEXT: vextracti128 $1, %ymm0, %xmm0
-; AVX512VL-NEXT: vpbroadcastq %xmm0, %zmm0
-; AVX512VL-NEXT: vpsllq $63, %zmm0, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,2,3,3]
+; AVX512VL-NEXT: vpbroadcastq %xmm1, %ymm1
+; AVX512VL-NEXT: vpslld $31, %ymm1, %ymm1
+; AVX512VL-NEXT: vptestmd %ymm1, %ymm1, %k1
; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; AVX512VL-NEXT: vpmovdw %ymm0, %xmm0
; AVX512VL-NEXT: vzeroupper
@@ -264,10 +262,10 @@ define <8 x i1> @shuf8i1_u_2_u_u_2_u_2_u(i8 %a) {
; VL_BW_DQ-LABEL: shuf8i1_u_2_u_u_2_u_2_u:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: kmovd %edi, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT: vextracti128 $1, %ymm0, %xmm0
-; VL_BW_DQ-NEXT: vpbroadcastq %xmm0, %zmm0
-; VL_BW_DQ-NEXT: vpmovq2m %zmm0, %k0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,2,3,3]
+; VL_BW_DQ-NEXT: vpbroadcastq %xmm0, %ymm0
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: vpmovm2w %k0, %xmm0
; VL_BW_DQ-NEXT: vzeroupper
; VL_BW_DQ-NEXT: retq
@@ -294,12 +292,14 @@ define i8 @shuf8i1_10_2_9_u_3_u_2_u(i8 %a) {
; AVX512VL-LABEL: shuf8i1_10_2_9_u_3_u_2_u:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: kmovw %edi, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
+; AVX512VL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = <8,2,10,u,3,u,2,u>
-; AVX512VL-NEXT: vpermi2q %zmm1, %zmm0, %zmm2
-; AVX512VL-NEXT: vpsllq $63, %zmm2, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512VL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3,4,5,6,7]
+; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k0
; AVX512VL-NEXT: kmovw %k0, %eax
; AVX512VL-NEXT: # kill: def %al killed %al killed %eax
; AVX512VL-NEXT: vzeroupper
@@ -308,11 +308,12 @@ define i8 @shuf8i1_10_2_9_u_3_u_2_u(i8 %a) {
; VL_BW_DQ-LABEL: shuf8i1_10_2_9_u_3_u_2_u:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: kmovd %edi, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[3,2,2,3]
+; VL_BW_DQ-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,0,1]
; VL_BW_DQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm2 = <8,2,10,u,3,u,2,u>
-; VL_BW_DQ-NEXT: vpermi2q %zmm1, %zmm0, %zmm2
-; VL_BW_DQ-NEXT: vpmovq2m %zmm2, %k0
+; VL_BW_DQ-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3,4,5,6,7]
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: kmovd %k0, %eax
; VL_BW_DQ-NEXT: # kill: def %al killed %al killed %eax
; VL_BW_DQ-NEXT: vzeroupper
@@ -339,10 +340,11 @@ define i8 @shuf8i1_0_1_4_5_u_u_u_u(i8 %a) {
; AVX512VL-LABEL: shuf8i1_0_1_4_5_u_u_u_u:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: kmovw %edi, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512VL-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5,0,1,0,1]
-; AVX512VL-NEXT: vpsllq $63, %zmm0, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k0
; AVX512VL-NEXT: kmovw %k0, %eax
; AVX512VL-NEXT: # kill: def %al killed %al killed %eax
; AVX512VL-NEXT: vzeroupper
@@ -351,9 +353,9 @@ define i8 @shuf8i1_0_1_4_5_u_u_u_u(i8 %a) {
; VL_BW_DQ-LABEL: shuf8i1_0_1_4_5_u_u_u_u:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: kmovd %edi, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT: vshufi64x2 {{.*#+}} zmm0 = zmm0[0,1,4,5,0,1,0,1]
-; VL_BW_DQ-NEXT: vpmovq2m %zmm0, %k0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: kmovd %k0, %eax
; VL_BW_DQ-NEXT: # kill: def %al killed %al killed %eax
; VL_BW_DQ-NEXT: vzeroupper
@@ -382,12 +384,13 @@ define i8 @shuf8i1_9_6_1_0_3_7_7_0(i8 %a) {
; AVX512VL-LABEL: shuf8i1_9_6_1_0_3_7_7_0:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: kmovw %edi, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,6,1,0,3,7,7,0]
-; AVX512VL-NEXT: vpermi2q %zmm1, %zmm0, %zmm2
-; AVX512VL-NEXT: vpsllq $63, %zmm2, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [8,6,1,0,3,7,7,0]
+; AVX512VL-NEXT: vpermi2d %ymm1, %ymm0, %ymm2
+; AVX512VL-NEXT: vpslld $31, %ymm2, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k0
; AVX512VL-NEXT: kmovw %k0, %eax
; AVX512VL-NEXT: # kill: def %al killed %al killed %eax
; AVX512VL-NEXT: vzeroupper
@@ -396,11 +399,11 @@ define i8 @shuf8i1_9_6_1_0_3_7_7_0(i8 %a) {
; VL_BW_DQ-LABEL: shuf8i1_9_6_1_0_3_7_7_0:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: kmovd %edi, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
; VL_BW_DQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,6,1,0,3,7,7,0]
-; VL_BW_DQ-NEXT: vpermi2q %zmm1, %zmm0, %zmm2
-; VL_BW_DQ-NEXT: vpmovq2m %zmm2, %k0
+; VL_BW_DQ-NEXT: vmovdqa {{.*#+}} ymm2 = [8,6,1,0,3,7,7,0]
+; VL_BW_DQ-NEXT: vpermi2d %ymm1, %ymm0, %ymm2
+; VL_BW_DQ-NEXT: vpmovd2m %ymm2, %k0
; VL_BW_DQ-NEXT: kmovd %k0, %eax
; VL_BW_DQ-NEXT: # kill: def %al killed %al killed %eax
; VL_BW_DQ-NEXT: vzeroupper
@@ -429,12 +432,13 @@ define i8 @shuf8i1_9_6_1_10_3_7_7_0(i8 %a) {
; AVX512VL-LABEL: shuf8i1_9_6_1_10_3_7_7_0:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: kmovw %edi, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,1,2,10,4,5,6,7]
-; AVX512VL-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX512VL-NEXT: vpermt2q %zmm0, %zmm1, %zmm2
-; AVX512VL-NEXT: vpsllq $63, %zmm2, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
+; AVX512VL-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; AVX512VL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3],ymm1[4,5,6,7]
+; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k0
; AVX512VL-NEXT: kmovw %k0, %eax
; AVX512VL-NEXT: # kill: def %al killed %al killed %eax
; AVX512VL-NEXT: vzeroupper
@@ -443,11 +447,11 @@ define i8 @shuf8i1_9_6_1_10_3_7_7_0(i8 %a) {
; VL_BW_DQ-LABEL: shuf8i1_9_6_1_10_3_7_7_0:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: kmovd %edi, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,1,2,10,4,5,6,7]
-; VL_BW_DQ-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; VL_BW_DQ-NEXT: vpermt2q %zmm0, %zmm1, %zmm2
-; VL_BW_DQ-NEXT: vpmovq2m %zmm2, %k0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,2]
+; VL_BW_DQ-NEXT: vpxor %xmm1, %xmm1, %xmm1
+; VL_BW_DQ-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2],ymm0[3],ymm1[4,5,6,7]
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: kmovd %k0, %eax
; VL_BW_DQ-NEXT: # kill: def %al killed %al killed %eax
; VL_BW_DQ-NEXT: vzeroupper
@@ -462,12 +466,10 @@ define i8 @shuf8i1__9_6_1_10_3_7_7_1(i8 %a) {
; AVX512F-LABEL: shuf8i1__9_6_1_10_3_7_7_1:
; AVX512F: # %bb.0:
; AVX512F-NEXT: kmovw %edi, %k1
-; AVX512F-NEXT: movb $51, %al
-; AVX512F-NEXT: kmovw %eax, %k2
-; AVX512F-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
-; AVX512F-NEXT: vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [9,6,1,0,3,7,7,1]
-; AVX512F-NEXT: vpermi2q %zmm1, %zmm0, %zmm2
+; AVX512F-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,6,1,0,3,7,7,1]
+; AVX512F-NEXT: vmovdqa64 {{.*#+}} zmm2 = [18446744073709551615,18446744073709551615,0,0,18446744073709551615,18446744073709551615,0,0]
+; AVX512F-NEXT: vpermt2q %zmm0, %zmm1, %zmm2
; AVX512F-NEXT: vpsllq $63, %zmm2, %zmm0
; AVX512F-NEXT: vptestmq %zmm0, %zmm0, %k0
; AVX512F-NEXT: kmovw %k0, %eax
@@ -478,14 +480,12 @@ define i8 @shuf8i1__9_6_1_10_3_7_7_1(i8 %a) {
; AVX512VL-LABEL: shuf8i1__9_6_1_10_3_7_7_1:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: kmovw %edi, %k1
-; AVX512VL-NEXT: movb $51, %al
-; AVX512VL-NEXT: kmovw %eax, %k2
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k2} {z}
-; AVX512VL-NEXT: vpternlogq $255, %zmm1, %zmm1, %zmm1 {%k1} {z}
-; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm2 = [9,6,1,0,3,7,7,1]
-; AVX512VL-NEXT: vpermi2q %zmm1, %zmm0, %zmm2
-; AVX512VL-NEXT: vpsllq $63, %zmm2, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm0 {%k1} {z}
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; AVX512VL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],mem[1,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k0
; AVX512VL-NEXT: kmovw %k0, %eax
; AVX512VL-NEXT: # kill: def %al killed %al killed %eax
; AVX512VL-NEXT: vzeroupper
@@ -494,11 +494,10 @@ define i8 @shuf8i1__9_6_1_10_3_7_7_1(i8 %a) {
; VL_BW_DQ-LABEL: shuf8i1__9_6_1_10_3_7_7_1:
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: kmovd %edi, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,6,1,0,3,7,7,1]
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm2 = [18446744073709551615,18446744073709551615,0,0,18446744073709551615,18446744073709551615,0,0]
-; VL_BW_DQ-NEXT: vpermt2q %zmm0, %zmm1, %zmm2
-; VL_BW_DQ-NEXT: vpmovq2m %zmm2, %k0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; VL_BW_DQ-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],mem[1,2,3,4,5,6,7]
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: kmovd %k0, %eax
; VL_BW_DQ-NEXT: # kill: def %al killed %al killed %eax
; VL_BW_DQ-NEXT: vzeroupper
@@ -528,15 +527,15 @@ define i8 @shuf8i1_9_6_1_10_3_7_7_0_all_ones(<8 x i1> %a) {
;
; AVX512VL-LABEL: shuf8i1_9_6_1_10_3_7_7_0_all_ones:
; AVX512VL: # %bb.0:
-; AVX512VL-NEXT: vpmovsxwq %xmm0, %zmm0
-; AVX512VL-NEXT: vpsllq $63, %zmm0, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k1
-; AVX512VL-NEXT: vpternlogq $255, %zmm0, %zmm0, %zmm0 {%k1} {z}
-; AVX512VL-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,1,2,3,4,5,6,7]
-; AVX512VL-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; AVX512VL-NEXT: vpermt2q %zmm0, %zmm1, %zmm2
-; AVX512VL-NEXT: vpsllq $63, %zmm2, %zmm0
-; AVX512VL-NEXT: vptestmq %zmm0, %zmm0, %k0
+; AVX512VL-NEXT: vpmovsxwd %xmm0, %ymm0
+; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k1
+; AVX512VL-NEXT: vpcmpeqd %ymm0, %ymm0, %ymm0
+; AVX512VL-NEXT: vmovdqa32 %ymm0, %ymm1 {%k1} {z}
+; AVX512VL-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[1,1,2,3]
+; AVX512VL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7]
+; AVX512VL-NEXT: vpslld $31, %ymm0, %ymm0
+; AVX512VL-NEXT: vptestmd %ymm0, %ymm0, %k0
; AVX512VL-NEXT: kmovw %k0, %eax
; AVX512VL-NEXT: # kill: def %al killed %al killed %eax
; AVX512VL-NEXT: vzeroupper
@@ -546,11 +545,11 @@ define i8 @shuf8i1_9_6_1_10_3_7_7_0_all_ones(<8 x i1> %a) {
; VL_BW_DQ: # %bb.0:
; VL_BW_DQ-NEXT: vpsllw $15, %xmm0, %xmm0
; VL_BW_DQ-NEXT: vpmovw2m %xmm0, %k0
-; VL_BW_DQ-NEXT: vpmovm2q %k0, %zmm0
-; VL_BW_DQ-NEXT: vmovdqa64 {{.*#+}} zmm1 = [9,1,2,3,4,5,6,7]
-; VL_BW_DQ-NEXT: vpternlogd $255, %zmm2, %zmm2, %zmm2
-; VL_BW_DQ-NEXT: vpermt2q %zmm0, %zmm1, %zmm2
-; VL_BW_DQ-NEXT: vpmovq2m %zmm2, %k0
+; VL_BW_DQ-NEXT: vpmovm2d %k0, %ymm0
+; VL_BW_DQ-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[1,1,2,3]
+; VL_BW_DQ-NEXT: vpcmpeqd %ymm1, %ymm1, %ymm1
+; VL_BW_DQ-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
+; VL_BW_DQ-NEXT: vpmovd2m %ymm0, %k0
; VL_BW_DQ-NEXT: kmovd %k0, %eax
; VL_BW_DQ-NEXT: # kill: def %al killed %al killed %eax
; VL_BW_DQ-NEXT: vzeroupper
generated by cgit v1.2.3 (git 2.25.1) at 2025-10-11 02:13:58 +0000