summaryrefslogtreecommitdiffstats
path: root/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll')
-rw-r--r--llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll158
1 files changed, 110 insertions, 48 deletions
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
index e0c3a51db25..d268364fda8 100644
--- a/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
+++ b/llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
@@ -578,18 +578,22 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_16_
;
; AVX2-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX2: # %bb.0:
-; AVX2-NEXT: vpbroadcastb %xmm0, %ymm1
-; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,3,0,1]
-; AVX2-NEXT: vpslldq {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16]
+; AVX2-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX2-NEXT: vpxor %xmm2, %xmm2, %xmm2
+; AVX2-NEXT: vpshufb %ymm2, %ymm0, %ymm0
+; AVX2-NEXT: vpslldq {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16]
; AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,0]
-; AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0
+; AVX2-NEXT: vpblendvb %ymm2, %ymm0, %ymm1, %ymm0
; AVX2-NEXT: retq
;
; AVX512VLBW-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_16_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
+; AVX512VLBW-NEXT: vpxor %xmm2, %xmm2, %xmm2
+; AVX512VLBW-NEXT: vpshufb %ymm2, %ymm0, %ymm0
; AVX512VLBW-NEXT: vpslldq {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm1[16]
-; AVX512VLBW-NEXT: vpbroadcastb %xmm0, %ymm0
; AVX512VLBW-NEXT: movl $-2147450880, %eax # imm = 0x80008000
; AVX512VLBW-NEXT: kmovd %eax, %k1
; AVX512VLBW-NEXT: vmovdqu8 %ymm1, %ymm0 {%k1}
@@ -920,11 +924,18 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_00_24_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_24_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_24_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_24_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,0,8,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_00_00_00_00_00_00_24_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -952,11 +963,18 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_00_25_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,9,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_00_00_00_00_00_25_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,9,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_00_00_00_00_00_25_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,9,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_00_25_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,0,9,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_00_00_00_00_00_25_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -984,11 +1002,18 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_00_26_00_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_00_00_00_00_26_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_00_00_00_00_26_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_00_00_00_00_26_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,0,10,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_00_00_00_00_26_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -1016,11 +1041,18 @@ define <32 x i8> @shuffle_v32i8_00_00_00_00_27_00_00_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_00_00_00_27_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_00_00_00_27_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_00_00_00_27_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,0,11,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_00_00_00_27_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -1048,11 +1080,18 @@ define <32 x i8> @shuffle_v32i8_00_00_00_28_00_00_00_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_00_00_28_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_00_00_28_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_00_00_28_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,0,12,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_00_00_28_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -1080,11 +1119,18 @@ define <32 x i8> @shuffle_v32i8_00_00_29_00_00_00_00_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_00_29_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_00_29_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_00_29_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,0,13,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_00_29_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -1112,11 +1158,18 @@ define <32 x i8> @shuffle_v32i8_00_30_00_00_00_00_00_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_00_30_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_00_30_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_00_30_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_00_30_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
@@ -1146,13 +1199,22 @@ define <32 x i8> @shuffle_v32i8_31_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_
; AVX2-NEXT: vpshufb %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
;
-; AVX512VLBW-LABEL: shuffle_v32i8_31_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
-; AVX512VLBW: # %bb.0:
-; AVX512VLBW-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
-; AVX512VLBW-NEXT: movl $15, %eax
-; AVX512VLBW-NEXT: vmovd %eax, %xmm1
-; AVX512VLBW-NEXT: vpshufb %ymm1, %ymm0, %ymm0
-; AVX512VLBW-NEXT: retq
+; AVX512VLBW-SLOW-LABEL: shuffle_v32i8_31_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-SLOW: # %bb.0:
+; AVX512VLBW-SLOW-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,3,0,1]
+; AVX512VLBW-SLOW-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7]
+; AVX512VLBW-SLOW-NEXT: movl $15, %eax
+; AVX512VLBW-SLOW-NEXT: vmovd %eax, %xmm1
+; AVX512VLBW-SLOW-NEXT: vpshufb %ymm1, %ymm0, %ymm0
+; AVX512VLBW-SLOW-NEXT: retq
+;
+; AVX512VLBW-FAST-LABEL: shuffle_v32i8_31_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
+; AVX512VLBW-FAST: # %bb.0:
+; AVX512VLBW-FAST-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,1]
+; AVX512VLBW-FAST-NEXT: movl $15, %eax
+; AVX512VLBW-FAST-NEXT: vmovd %eax, %xmm1
+; AVX512VLBW-FAST-NEXT: vpshufb %ymm1, %ymm0, %ymm0
+; AVX512VLBW-FAST-NEXT: retq
;
; AVX512VLVBMI-LABEL: shuffle_v32i8_31_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00_00:
; AVX512VLVBMI: # %bb.0:
OpenPOWER on IntegriCloud