@@ -2011,15 +2011,14 @@ define void @vec384_i8_widen_to_i24_factor3_broadcast_to_v16i24_factor16(ptr %in
20112011;
20122012; AVX2-LABEL: vec384_i8_widen_to_i24_factor3_broadcast_to_v16i24_factor16:
20132013; AVX2: # %bb.0:
2014- ; AVX2-NEXT: vpbroadcastb (%rdi), %xmm0
2015- ; AVX2-NEXT: vmovdqa (%rdi), % xmm1
2016- ; AVX2-NEXT: vpshufb {{.*#+}} xmm2 = zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero
2017- ; AVX2-NEXT: vinserti128 $1, %xmm2 , %ymm0, %ymm0
2014+ ; AVX2-NEXT: vmovdqa (%rdi), %xmm0
2015+ ; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,xmm0[0],zero,zero,xmm0[0],zero,zero,xmm0[0],zero,zero,xmm0[0],zero,zero,xmm0[0],zero,zero
2016+ ; AVX2-NEXT: vpbroadcastb %xmm0, %ymm0
2017+ ; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip) , %ymm0, %ymm0
20182018; AVX2-NEXT: vperm2i128 {{.*#+}} ymm2 = mem[2,3],ymm0[2,3]
20192019; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0,0,255,255,0,255,255,0,255,255,0,255,255,0,255,255,0]
20202020; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
20212021; AVX2-NEXT: vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
2022- ; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero,xmm1[0],zero,zero
20232022; AVX2-NEXT: vpaddb (%rsi), %ymm0, %ymm0
20242023; AVX2-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
20252024; AVX2-NEXT: vmovdqa %ymm1, 32(%rdx)
@@ -2263,15 +2262,14 @@ define void @vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8(ptr %in.e
22632262;
22642263; AVX2-LABEL: vec384_i8_widen_to_i48_factor6_broadcast_to_v8i48_factor8:
22652264; AVX2: # %bb.0:
2266- ; AVX2-NEXT: vpbroadcastb (%rdi), %xmm0
2267- ; AVX2-NEXT: vmovdqa (%rdi), % xmm1
2268- ; AVX2-NEXT: vpshufb {{.*#+}} xmm2 = zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero
2269- ; AVX2-NEXT: vinserti128 $1, %xmm2 , %ymm0, %ymm0
2265+ ; AVX2-NEXT: vmovdqa (%rdi), %xmm0
2266+ ; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm0[0],zero,zero,zero,zero,zero,xmm0[0],zero,zero,zero,zero,zero
2267+ ; AVX2-NEXT: vpbroadcastb %xmm0, %ymm0
2268+ ; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip) , %ymm0, %ymm0
22702269; AVX2-NEXT: vperm2i128 {{.*#+}} ymm2 = mem[2,3],ymm0[2,3]
22712270; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,255,255,255,0,255,255,255,255,255,0,255,255,255,0,255,255,255,255,255,0,255,255,255,255,255,0,255,255,255]
22722271; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
22732272; AVX2-NEXT: vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
2274- ; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero
22752273; AVX2-NEXT: vpaddb (%rsi), %ymm0, %ymm0
22762274; AVX2-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
22772275; AVX2-NEXT: vmovdqa %ymm1, 32(%rdx)
@@ -2515,15 +2513,14 @@ define void @vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4(ptr %in.
25152513;
25162514; AVX2-LABEL: vec384_i8_widen_to_i96_factor12_broadcast_to_v4i96_factor4:
25172515; AVX2: # %bb.0:
2518- ; AVX2-NEXT: vpbroadcastb (%rdi), %xmm0
2519- ; AVX2-NEXT: vmovdqa (%rdi), % xmm1
2520- ; AVX2-NEXT: vpshufb {{.*#+}} xmm2 = zero,zero,zero,zero,zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero
2521- ; AVX2-NEXT: vinserti128 $1, %xmm2 , %ymm0, %ymm0
2516+ ; AVX2-NEXT: vmovdqa (%rdi), %xmm0
2517+ ; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm0[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
2518+ ; AVX2-NEXT: vpbroadcastb %xmm0, %ymm0
2519+ ; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip) , %ymm0, %ymm0
25222520; AVX2-NEXT: vperm2i128 {{.*#+}} ymm2 = mem[2,3],ymm0[2,3]
25232521; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [0,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255,0,255,255,255,255,255,255,255,255,255,255,255,0,255,255,255]
25242522; AVX2-NEXT: # ymm3 = mem[0,1,0,1]
25252523; AVX2-NEXT: vpblendvb %ymm3, %ymm2, %ymm0, %ymm0
2526- ; AVX2-NEXT: vpshufb {{.*#+}} xmm1 = zero,zero,zero,zero,xmm1[0],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero
25272524; AVX2-NEXT: vpaddb (%rsi), %ymm0, %ymm0
25282525; AVX2-NEXT: vpaddb 32(%rsi), %ymm1, %ymm1
25292526; AVX2-NEXT: vmovdqa %ymm1, 32(%rdx)
0 commit comments