1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=SSE2 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2 4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512,AVX512F 5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefixes=AVX,AVX512,AVX512BW 6 7define <4 x i32> @test1(<8 x i32> %v) { 8; SSE2-LABEL: test1: 9; SSE2: # %bb.0: 10; SSE2-NEXT: retq 11; 12; AVX-LABEL: test1: 13; AVX: # %bb.0: 14; AVX-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 15; AVX-NEXT: vzeroupper 16; AVX-NEXT: retq 17 %x = sext <8 x i32> %v to <8 x i64> 18 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 19 %t = trunc <4 x i64> %s to <4 x i32> 20 ret <4 x i32> %t 21} 22 23define <4 x i32> @test2(<8 x i32> %v) { 24; SSE2-LABEL: test2: 25; SSE2: # %bb.0: 26; SSE2-NEXT: movaps %xmm1, %xmm0 27; SSE2-NEXT: retq 28; 29; AVX-LABEL: test2: 30; AVX: # %bb.0: 31; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0 32; AVX-NEXT: vzeroupper 33; AVX-NEXT: retq 34 %x = sext <8 x i32> %v to <8 x i64> 35 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 36 %t = trunc <4 x i64> %s to <4 x i32> 37 ret <4 x i32> %t 38} 39 40define <2 x i32> @test3(<8 x i32> %v) { 41; SSE2-LABEL: test3: 42; SSE2: # %bb.0: 43; SSE2-NEXT: movaps %xmm1, %xmm0 44; SSE2-NEXT: retq 45; 46; AVX-LABEL: test3: 47; AVX: # %bb.0: 48; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0 49; AVX-NEXT: vzeroupper 50; AVX-NEXT: retq 51 %x = sext <8 x i32> %v to <8 x i64> 52 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 4, i32 5> 53 %t = trunc <2 x i64> %s to <2 x i32> 54 ret <2 x i32> %t 55} 56 57define <2 x i32> @test4(<8 x i32> %v) { 58; SSE2-LABEL: test4: 59; SSE2: # %bb.0: 60; SSE2-NEXT: retq 61; 62; AVX-LABEL: test4: 63; AVX: # %bb.0: 64; AVX-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 65; AVX-NEXT: vzeroupper 66; AVX-NEXT: retq 67 %x = sext <8 x i32> %v to <8 x i64> 68 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 0, i32 1> 69 %t = trunc <2 x i64> %s to <2 x i32> 70 ret <2 x i32> %t 71} 72 73define <2 x i32> @test5(<8 x i32> %v) { 74; SSE2-LABEL: test5: 75; SSE2: # %bb.0: 76; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,2,2] 77; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 78; SSE2-NEXT: retq 79; 80; AVX2-LABEL: test5: 81; AVX2: # %bb.0: 82; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 83; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 84; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 85; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7] 86; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm1 = [6,6,6,6] 87; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0 88; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 89; AVX2-NEXT: vzeroupper 90; AVX2-NEXT: retq 91; 92; AVX512-LABEL: test5: 93; AVX512: # %bb.0: 94; AVX512-NEXT: vpmovsxdq %ymm0, %zmm0 95; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1 96; AVX512-NEXT: vpextrq $1, %xmm1, %rax 97; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm0 98; AVX512-NEXT: vmovq %xmm0, %rcx 99; AVX512-NEXT: vmovd %eax, %xmm0 100; AVX512-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0 101; AVX512-NEXT: vzeroupper 102; AVX512-NEXT: retq 103 %x = sext <8 x i32> %v to <8 x i64> 104 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 3, i32 4> 105 %t = trunc <2 x i64> %s to <2 x i32> 106 ret <2 x i32> %t 107} 108 109define <4 x i32> @test6(<8 x i32> %v) { 110; SSE2-LABEL: test6: 111; SSE2: # %bb.0: 112; SSE2-NEXT: retq 113; 114; AVX-LABEL: test6: 115; AVX: # %bb.0: 116; AVX-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 117; AVX-NEXT: vzeroupper 118; AVX-NEXT: retq 119 %x = zext <8 x i32> %v to <8 x i64> 120 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 121 %t = trunc <4 x i64> %s to <4 x i32> 122 ret <4 x i32> %t 123} 124 125define <4 x i32> @test7(<8 x i32> %v) { 126; SSE2-LABEL: test7: 127; SSE2: # %bb.0: 128; SSE2-NEXT: movaps %xmm1, %xmm0 129; SSE2-NEXT: retq 130; 131; AVX-LABEL: test7: 132; AVX: # %bb.0: 133; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0 134; AVX-NEXT: vzeroupper 135; AVX-NEXT: retq 136 %x = zext <8 x i32> %v to <8 x i64> 137 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 138 %t = trunc <4 x i64> %s to <4 x i32> 139 ret <4 x i32> %t 140} 141 142define <2 x i32> @test8(<8 x i32> %v) { 143; SSE2-LABEL: test8: 144; SSE2: # %bb.0: 145; SSE2-NEXT: movaps %xmm1, %xmm0 146; SSE2-NEXT: retq 147; 148; AVX-LABEL: test8: 149; AVX: # %bb.0: 150; AVX-NEXT: vextractf128 $1, %ymm0, %xmm0 151; AVX-NEXT: vzeroupper 152; AVX-NEXT: retq 153 %x = zext <8 x i32> %v to <8 x i64> 154 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 4, i32 5> 155 %t = trunc <2 x i64> %s to <2 x i32> 156 ret <2 x i32> %t 157} 158 159define <2 x i32> @test9(<8 x i32> %v) { 160; SSE2-LABEL: test9: 161; SSE2: # %bb.0: 162; SSE2-NEXT: retq 163; 164; AVX-LABEL: test9: 165; AVX: # %bb.0: 166; AVX-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 167; AVX-NEXT: vzeroupper 168; AVX-NEXT: retq 169 %x = zext <8 x i32> %v to <8 x i64> 170 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 0, i32 1> 171 %t = trunc <2 x i64> %s to <2 x i32> 172 ret <2 x i32> %t 173} 174 175define <2 x i32> @test10(<8 x i32> %v) { 176; SSE2-LABEL: test10: 177; SSE2: # %bb.0: 178; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[3,3,2,2] 179; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 180; SSE2-NEXT: retq 181; 182; AVX2-LABEL: test10: 183; AVX2: # %bb.0: 184; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm1 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 185; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm0 186; AVX2-NEXT: vpmovzxdq {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero 187; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7] 188; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm1 = [6,6,6,6] 189; AVX2-NEXT: vpermd %ymm0, %ymm1, %ymm0 190; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 191; AVX2-NEXT: vzeroupper 192; AVX2-NEXT: retq 193; 194; AVX512-LABEL: test10: 195; AVX512: # %bb.0: 196; AVX512-NEXT: vpmovzxdq {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero 197; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm1 198; AVX512-NEXT: vpextrq $1, %xmm1, %rax 199; AVX512-NEXT: vextracti32x4 $2, %zmm0, %xmm0 200; AVX512-NEXT: vmovq %xmm0, %rcx 201; AVX512-NEXT: vmovd %eax, %xmm0 202; AVX512-NEXT: vpinsrd $1, %ecx, %xmm0, %xmm0 203; AVX512-NEXT: vzeroupper 204; AVX512-NEXT: retq 205 %x = zext <8 x i32> %v to <8 x i64> 206 %s = shufflevector <8 x i64> %x, <8 x i64> undef, <2 x i32> <i32 3, i32 4> 207 %t = trunc <2 x i64> %s to <2 x i32> 208 ret <2 x i32> %t 209} 210