1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-- -mattr=+avx | FileCheck %s --check-prefixes=CHECK,X86 3; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,X64 4 5; PR 9267 6 7define <4 x i16> @func_16_32(ptr %a, ptr %b, ptr %c) nounwind { 8; X86-LABEL: func_16_32: 9; X86: # %bb.0: 10; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 11; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 12; X86-NEXT: movl {{[0-9]+}}(%esp), %edx 13; X86-NEXT: vmovdqa (%edx), %xmm0 14; X86-NEXT: vpaddw (%ecx), %xmm0, %xmm0 15; X86-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15] 16; X86-NEXT: vmovq %xmm0, (%eax) 17; X86-NEXT: retl 18; 19; X64-LABEL: func_16_32: 20; X64: # %bb.0: 21; X64-NEXT: vmovdqa (%rsi), %xmm0 22; X64-NEXT: vpaddw (%rdi), %xmm0, %xmm0 23; X64-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15] 24; X64-NEXT: vmovq %xmm0, (%rdx) 25; X64-NEXT: retq 26 %F = load <4 x i32>, ptr %a 27 %G = trunc <4 x i32> %F to <4 x i16> 28 %H = load <4 x i32>, ptr %b 29 %Y = trunc <4 x i32> %H to <4 x i16> 30 %T = add <4 x i16> %Y, %G 31 store <4 x i16>%T , ptr %c 32 ret <4 x i16> %T 33} 34 35define <4 x i16> @func_16_64(ptr %a, ptr %b, ptr %c) nounwind { 36; X86-LABEL: func_16_64: 37; X86: # %bb.0: 38; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 39; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 40; X86-NEXT: movl {{[0-9]+}}(%esp), %edx 41; X86-NEXT: vmovaps (%edx), %ymm0 42; X86-NEXT: vxorps (%ecx), %ymm0, %ymm0 43; X86-NEXT: vextractf128 $1, %ymm0, %xmm1 44; X86-NEXT: vpxor %xmm2, %xmm2, %xmm2 45; X86-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1,2,3],xmm1[4],xmm2[5,6,7] 46; X86-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1,2,3],xmm0[4],xmm2[5,6,7] 47; X86-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 48; X86-NEXT: vpackusdw %xmm0, %xmm0, %xmm0 49; X86-NEXT: vmovq %xmm0, (%eax) 50; X86-NEXT: vzeroupper 51; X86-NEXT: retl 52; 53; X64-LABEL: func_16_64: 54; X64: # %bb.0: 55; X64-NEXT: vmovdqa (%rsi), %ymm0 56; X64-NEXT: vpxor (%rdi), %ymm0, %ymm0 57; X64-NEXT: vpxor %xmm1, %xmm1, %xmm1 58; X64-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3],ymm0[4],ymm1[5,6,7],ymm0[8],ymm1[9,10,11],ymm0[12],ymm1[13,14,15] 59; X64-NEXT: vextracti128 $1, %ymm0, %xmm1 60; X64-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 61; X64-NEXT: vpackusdw %xmm0, %xmm0, %xmm0 62; X64-NEXT: vmovq %xmm0, (%rdx) 63; X64-NEXT: vzeroupper 64; X64-NEXT: retq 65 %F = load <4 x i64>, ptr %a 66 %G = trunc <4 x i64> %F to <4 x i16> 67 %H = load <4 x i64>, ptr %b 68 %Y = trunc <4 x i64> %H to <4 x i16> 69 %T = xor <4 x i16> %Y, %G 70 store <4 x i16>%T , ptr %c 71 ret <4 x i16> %T 72} 73 74define <4 x i32> @func_32_64(ptr %a, ptr %b) nounwind { 75; X86-LABEL: func_32_64: 76; X86: # %bb.0: 77; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 78; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 79; X86-NEXT: vmovaps (%ecx), %ymm0 80; X86-NEXT: vorps (%eax), %ymm0, %ymm0 81; X86-NEXT: vextractf128 $1, %ymm0, %xmm1 82; X86-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2] 83; X86-NEXT: vzeroupper 84; X86-NEXT: retl 85; 86; X64-LABEL: func_32_64: 87; X64: # %bb.0: 88; X64-NEXT: vmovaps (%rsi), %ymm0 89; X64-NEXT: vorps (%rdi), %ymm0, %ymm0 90; X64-NEXT: vextractf128 $1, %ymm0, %xmm1 91; X64-NEXT: vshufps {{.*#+}} xmm0 = xmm0[0,2],xmm1[0,2] 92; X64-NEXT: vzeroupper 93; X64-NEXT: retq 94 %F = load <4 x i64>, ptr %a 95 %G = trunc <4 x i64> %F to <4 x i32> 96 %H = load <4 x i64>, ptr %b 97 %Y = trunc <4 x i64> %H to <4 x i32> 98 %T = or <4 x i32> %Y, %G 99 ret <4 x i32> %T 100} 101 102define <4 x i8> @func_8_16(ptr %a, ptr %b) nounwind { 103; X86-LABEL: func_8_16: 104; X86: # %bb.0: 105; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 106; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 107; X86-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 108; X86-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero 109; X86-NEXT: vpaddb %xmm0, %xmm1, %xmm0 110; X86-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,u,u,u,u,u,u,u,u,u,u,u,u] 111; X86-NEXT: retl 112; 113; X64-LABEL: func_8_16: 114; X64: # %bb.0: 115; X64-NEXT: vmovq {{.*#+}} xmm0 = mem[0],zero 116; X64-NEXT: vmovq {{.*#+}} xmm1 = mem[0],zero 117; X64-NEXT: vpaddb %xmm0, %xmm1, %xmm0 118; X64-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,u,u,u,u,u,u,u,u,u,u,u,u] 119; X64-NEXT: retq 120 %F = load <4 x i16>, ptr %a 121 %G = trunc <4 x i16> %F to <4 x i8> 122 %H = load <4 x i16>, ptr %b 123 %Y = trunc <4 x i16> %H to <4 x i8> 124 %T = add <4 x i8> %Y, %G 125 ret <4 x i8> %T 126} 127 128define <4 x i8> @func_8_32(ptr %a, ptr %b) nounwind { 129; X86-LABEL: func_8_32: 130; X86: # %bb.0: 131; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 132; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 133; X86-NEXT: vmovdqa (%ecx), %xmm0 134; X86-NEXT: vpsubb (%eax), %xmm0, %xmm0 135; X86-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u] 136; X86-NEXT: retl 137; 138; X64-LABEL: func_8_32: 139; X64: # %bb.0: 140; X64-NEXT: vmovdqa (%rsi), %xmm0 141; X64-NEXT: vpsubb (%rdi), %xmm0, %xmm0 142; X64-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u] 143; X64-NEXT: retq 144 %F = load <4 x i32>, ptr %a 145 %G = trunc <4 x i32> %F to <4 x i8> 146 %H = load <4 x i32>, ptr %b 147 %Y = trunc <4 x i32> %H to <4 x i8> 148 %T = sub <4 x i8> %Y, %G 149 ret <4 x i8> %T 150} 151 152define <4 x i8> @func_8_64(ptr %a, ptr %b) nounwind { 153; X86-LABEL: func_8_64: 154; X86: # %bb.0: 155; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 156; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 157; X86-NEXT: vmovdqa (%ecx), %xmm0 158; X86-NEXT: vmovdqa 16(%ecx), %xmm1 159; X86-NEXT: vmovd {{.*#+}} xmm2 = [0,8,0,0,0,0,0,0,0,0,0,0,0,0,0,0] 160; X86-NEXT: vpshufb %xmm2, %xmm1, %xmm1 161; X86-NEXT: vpshufb %xmm2, %xmm0, %xmm0 162; X86-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 163; X86-NEXT: vmovdqa (%eax), %xmm1 164; X86-NEXT: vmovdqa 16(%eax), %xmm3 165; X86-NEXT: vpshufb %xmm2, %xmm3, %xmm3 166; X86-NEXT: vpshufb %xmm2, %xmm1, %xmm1 167; X86-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3] 168; X86-NEXT: vpaddb %xmm0, %xmm1, %xmm0 169; X86-NEXT: retl 170; 171; X64-LABEL: func_8_64: 172; X64: # %bb.0: 173; X64-NEXT: vmovdqa (%rdi), %xmm0 174; X64-NEXT: vmovdqa 16(%rdi), %xmm1 175; X64-NEXT: vpbroadcastw {{.*#+}} xmm2 = [0,8,0,8,0,8,0,8,0,8,0,8,0,8,0,8] 176; X64-NEXT: vpshufb %xmm2, %xmm1, %xmm1 177; X64-NEXT: vpshufb %xmm2, %xmm0, %xmm0 178; X64-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] 179; X64-NEXT: vmovdqa (%rsi), %xmm1 180; X64-NEXT: vmovdqa 16(%rsi), %xmm3 181; X64-NEXT: vpshufb %xmm2, %xmm3, %xmm3 182; X64-NEXT: vpshufb %xmm2, %xmm1, %xmm1 183; X64-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1],xmm1[2],xmm3[2],xmm1[3],xmm3[3] 184; X64-NEXT: vpaddb %xmm0, %xmm1, %xmm0 185; X64-NEXT: retq 186 %F = load <4 x i64>, ptr %a 187 %G = trunc <4 x i64> %F to <4 x i8> 188 %H = load <4 x i64>, ptr %b 189 %Y = trunc <4 x i64> %H to <4 x i8> 190 %T = add <4 x i8> %Y, %G 191 ret <4 x i8> %T 192} 193 194define <4 x i16> @const_16_32() nounwind { 195; CHECK-LABEL: const_16_32: 196; CHECK: # %bb.0: 197; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = [0,3,8,7,0,0,0,0] 198; CHECK-NEXT: ret{{[l|q]}} 199 %G = trunc <4 x i32> <i32 0, i32 3, i32 8, i32 7> to <4 x i16> 200 ret <4 x i16> %G 201} 202 203define <4 x i16> @const_16_64() nounwind { 204; CHECK-LABEL: const_16_64: 205; CHECK: # %bb.0: 206; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = [0,3,8,7,0,0,0,0] 207; CHECK-NEXT: ret{{[l|q]}} 208 %G = trunc <4 x i64> <i64 0, i64 3, i64 8, i64 7> to <4 x i16> 209 ret <4 x i16> %G 210} 211 212define void @bugOnTruncBitwidthReduce() nounwind { 213; CHECK-LABEL: bugOnTruncBitwidthReduce: 214; CHECK: # %bb.0: # %meh 215; CHECK-NEXT: ret{{[l|q]}} 216meh: 217 %0 = xor <4 x i64> zeroinitializer, zeroinitializer 218 %1 = trunc <4 x i64> %0 to <4 x i32> 219 %2 = lshr <4 x i32> %1, <i32 18, i32 18, i32 18, i32 18> 220 %3 = xor <4 x i32> %2, %1 221 ret void 222} 223