1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 | FileCheck %s --check-prefix=SSE2 3; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 | FileCheck %s --check-prefix=AVX2 4 5define void @trunc_shl_7_v4i32_v4i64(<4 x i32> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) { 6; SSE2-LABEL: trunc_shl_7_v4i32_v4i64: 7; SSE2: # %bb.0: 8; SSE2-NEXT: movaps (%rsi), %xmm0 9; SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[0,2],mem[0,2] 10; SSE2-NEXT: pslld $7, %xmm0 11; SSE2-NEXT: movdqa %xmm0, (%rdi) 12; SSE2-NEXT: retq 13; 14; AVX2-LABEL: trunc_shl_7_v4i32_v4i64: 15; AVX2: # %bb.0: 16; AVX2-NEXT: vpshufd {{.*#+}} ymm0 = mem[0,2,2,3,4,6,6,7] 17; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3] 18; AVX2-NEXT: vpslld $7, %xmm0, %xmm0 19; AVX2-NEXT: vmovdqa %xmm0, (%rdi) 20; AVX2-NEXT: vzeroupper 21; AVX2-NEXT: retq 22 %val = load <4 x i64>, <4 x i64> addrspace(1)* %in 23 %shl = shl <4 x i64> %val, <i64 7, i64 7, i64 7, i64 7> 24 %trunc = trunc <4 x i64> %shl to <4 x i32> 25 store <4 x i32> %trunc, <4 x i32> addrspace(1)* %out 26 ret void 27} 28 29define <8 x i16> @trunc_shl_15_v8i16_v8i32(<8 x i32> %a) { 30; SSE2-LABEL: trunc_shl_15_v8i16_v8i32: 31; SSE2: # %bb.0: 32; SSE2-NEXT: pslld $16, %xmm1 33; SSE2-NEXT: psrad $16, %xmm1 34; SSE2-NEXT: pslld $16, %xmm0 35; SSE2-NEXT: psrad $16, %xmm0 36; SSE2-NEXT: packssdw %xmm1, %xmm0 37; SSE2-NEXT: psllw $15, %xmm0 38; SSE2-NEXT: retq 39; 40; AVX2-LABEL: trunc_shl_15_v8i16_v8i32: 41; AVX2: # %bb.0: 42; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31] 43; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3] 44; AVX2-NEXT: vpsllw $15, %xmm0, %xmm0 45; AVX2-NEXT: vzeroupper 46; AVX2-NEXT: retq 47 %shl = shl <8 x i32> %a, <i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15> 48 %conv = trunc <8 x i32> %shl to <8 x i16> 49 ret <8 x i16> %conv 50} 51 52define <8 x i16> @trunc_shl_16_v8i16_v8i32(<8 x i32> %a) { 53; SSE2-LABEL: trunc_shl_16_v8i16_v8i32: 54; SSE2: # %bb.0: 55; SSE2-NEXT: xorps %xmm0, %xmm0 56; SSE2-NEXT: retq 57; 58; AVX2-LABEL: trunc_shl_16_v8i16_v8i32: 59; AVX2: # %bb.0: 60; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm0[28,29] 61; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 62; AVX2-NEXT: vzeroupper 63; AVX2-NEXT: retq 64 %shl = shl <8 x i32> %a, <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16> 65 %conv = trunc <8 x i32> %shl to <8 x i16> 66 ret <8 x i16> %conv 67} 68 69define <8 x i16> @trunc_shl_17_v8i16_v8i32(<8 x i32> %a) { 70; SSE2-LABEL: trunc_shl_17_v8i16_v8i32: 71; SSE2: # %bb.0: 72; SSE2-NEXT: xorps %xmm0, %xmm0 73; SSE2-NEXT: retq 74; 75; AVX2-LABEL: trunc_shl_17_v8i16_v8i32: 76; AVX2: # %bb.0: 77; AVX2-NEXT: vpslld $17, %ymm0, %ymm0 78; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15,16,17,20,21,24,25,28,29,24,25,28,29,28,29,30,31] 79; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3] 80; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 81; AVX2-NEXT: vzeroupper 82; AVX2-NEXT: retq 83 %shl = shl <8 x i32> %a, <i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17> 84 %conv = trunc <8 x i32> %shl to <8 x i16> 85 ret <8 x i16> %conv 86} 87 88define void @trunc_shl_31_i32_i64(i32* %out, i64* %in) { 89; SSE2-LABEL: trunc_shl_31_i32_i64: 90; SSE2: # %bb.0: 91; SSE2-NEXT: movl (%rsi), %eax 92; SSE2-NEXT: shll $31, %eax 93; SSE2-NEXT: movl %eax, (%rdi) 94; SSE2-NEXT: retq 95; 96; AVX2-LABEL: trunc_shl_31_i32_i64: 97; AVX2: # %bb.0: 98; AVX2-NEXT: movl (%rsi), %eax 99; AVX2-NEXT: shll $31, %eax 100; AVX2-NEXT: movl %eax, (%rdi) 101; AVX2-NEXT: retq 102 %val = load i64, i64* %in 103 %shl = shl i64 %val, 31 104 %trunc = trunc i64 %shl to i32 105 store i32 %trunc, i32* %out 106 ret void 107} 108 109define void @trunc_shl_32_i32_i64(i32* %out, i64* %in) { 110; SSE2-LABEL: trunc_shl_32_i32_i64: 111; SSE2: # %bb.0: 112; SSE2-NEXT: movl $0, (%rdi) 113; SSE2-NEXT: retq 114; 115; AVX2-LABEL: trunc_shl_32_i32_i64: 116; AVX2: # %bb.0: 117; AVX2-NEXT: movl $0, (%rdi) 118; AVX2-NEXT: retq 119 %val = load i64, i64* %in 120 %shl = shl i64 %val, 32 121 %trunc = trunc i64 %shl to i32 122 store i32 %trunc, i32* %out 123 ret void 124} 125 126define void @trunc_shl_15_i16_i64(i16* %out, i64* %in) { 127; SSE2-LABEL: trunc_shl_15_i16_i64: 128; SSE2: # %bb.0: 129; SSE2-NEXT: movl (%rsi), %eax 130; SSE2-NEXT: shll $15, %eax 131; SSE2-NEXT: movw %ax, (%rdi) 132; SSE2-NEXT: retq 133; 134; AVX2-LABEL: trunc_shl_15_i16_i64: 135; AVX2: # %bb.0: 136; AVX2-NEXT: movl (%rsi), %eax 137; AVX2-NEXT: shll $15, %eax 138; AVX2-NEXT: movw %ax, (%rdi) 139; AVX2-NEXT: retq 140 %val = load i64, i64* %in 141 %shl = shl i64 %val, 15 142 %trunc = trunc i64 %shl to i16 143 store i16 %trunc, i16* %out 144 ret void 145} 146 147define void @trunc_shl_16_i16_i64(i16* %out, i64* %in) { 148; SSE2-LABEL: trunc_shl_16_i16_i64: 149; SSE2: # %bb.0: 150; SSE2-NEXT: movw $0, (%rdi) 151; SSE2-NEXT: retq 152; 153; AVX2-LABEL: trunc_shl_16_i16_i64: 154; AVX2: # %bb.0: 155; AVX2-NEXT: movw $0, (%rdi) 156; AVX2-NEXT: retq 157 %val = load i64, i64* %in 158 %shl = shl i64 %val, 16 159 %trunc = trunc i64 %shl to i16 160 store i16 %trunc, i16* %out 161 ret void 162} 163 164define void @trunc_shl_7_i8_i64(i8* %out, i64* %in) { 165; SSE2-LABEL: trunc_shl_7_i8_i64: 166; SSE2: # %bb.0: 167; SSE2-NEXT: movb (%rsi), %al 168; SSE2-NEXT: shlb $7, %al 169; SSE2-NEXT: movb %al, (%rdi) 170; SSE2-NEXT: retq 171; 172; AVX2-LABEL: trunc_shl_7_i8_i64: 173; AVX2: # %bb.0: 174; AVX2-NEXT: movb (%rsi), %al 175; AVX2-NEXT: shlb $7, %al 176; AVX2-NEXT: movb %al, (%rdi) 177; AVX2-NEXT: retq 178 %val = load i64, i64* %in 179 %shl = shl i64 %val, 7 180 %trunc = trunc i64 %shl to i8 181 store i8 %trunc, i8* %out 182 ret void 183} 184 185define void @trunc_shl_8_i8_i64(i8* %out, i64* %in) { 186; SSE2-LABEL: trunc_shl_8_i8_i64: 187; SSE2: # %bb.0: 188; SSE2-NEXT: movb $0, (%rdi) 189; SSE2-NEXT: retq 190; 191; AVX2-LABEL: trunc_shl_8_i8_i64: 192; AVX2: # %bb.0: 193; AVX2-NEXT: movb $0, (%rdi) 194; AVX2-NEXT: retq 195 %val = load i64, i64* %in 196 %shl = shl i64 %val, 8 197 %trunc = trunc i64 %shl to i8 198 store i8 %trunc, i8* %out 199 ret void 200} 201