1; RUN: llc -mtriple=arm-eabi -mattr=+neon %s -o - | FileCheck %s 2 3define <8 x i8> @vsubi8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 4;CHECK-LABEL: vsubi8: 5;CHECK: vsub.i8 6 %tmp1 = load <8 x i8>, <8 x i8>* %A 7 %tmp2 = load <8 x i8>, <8 x i8>* %B 8 %tmp3 = sub <8 x i8> %tmp1, %tmp2 9 ret <8 x i8> %tmp3 10} 11 12define <4 x i16> @vsubi16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 13;CHECK-LABEL: vsubi16: 14;CHECK: vsub.i16 15 %tmp1 = load <4 x i16>, <4 x i16>* %A 16 %tmp2 = load <4 x i16>, <4 x i16>* %B 17 %tmp3 = sub <4 x i16> %tmp1, %tmp2 18 ret <4 x i16> %tmp3 19} 20 21define <2 x i32> @vsubi32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 22;CHECK-LABEL: vsubi32: 23;CHECK: vsub.i32 24 %tmp1 = load <2 x i32>, <2 x i32>* %A 25 %tmp2 = load <2 x i32>, <2 x i32>* %B 26 %tmp3 = sub <2 x i32> %tmp1, %tmp2 27 ret <2 x i32> %tmp3 28} 29 30define <1 x i64> @vsubi64(<1 x i64>* %A, <1 x i64>* %B) nounwind { 31;CHECK-LABEL: vsubi64: 32;CHECK: vsub.i64 33 %tmp1 = load <1 x i64>, <1 x i64>* %A 34 %tmp2 = load <1 x i64>, <1 x i64>* %B 35 %tmp3 = sub <1 x i64> %tmp1, %tmp2 36 ret <1 x i64> %tmp3 37} 38 39define <2 x float> @vsubf32(<2 x float>* %A, <2 x float>* %B) nounwind { 40;CHECK-LABEL: vsubf32: 41;CHECK: vsub.f32 42 %tmp1 = load <2 x float>, <2 x float>* %A 43 %tmp2 = load <2 x float>, <2 x float>* %B 44 %tmp3 = fsub <2 x float> %tmp1, %tmp2 45 ret <2 x float> %tmp3 46} 47 48define <16 x i8> @vsubQi8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 49;CHECK-LABEL: vsubQi8: 50;CHECK: vsub.i8 51 %tmp1 = load <16 x i8>, <16 x i8>* %A 52 %tmp2 = load <16 x i8>, <16 x i8>* %B 53 %tmp3 = sub <16 x i8> %tmp1, %tmp2 54 ret <16 x i8> %tmp3 55} 56 57define <8 x i16> @vsubQi16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 58;CHECK-LABEL: vsubQi16: 59;CHECK: vsub.i16 60 %tmp1 = load <8 x i16>, <8 x i16>* %A 61 %tmp2 = load <8 x i16>, <8 x i16>* %B 62 %tmp3 = sub <8 x i16> %tmp1, %tmp2 63 ret <8 x i16> %tmp3 64} 65 66define <4 x i32> @vsubQi32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 67;CHECK-LABEL: vsubQi32: 68;CHECK: vsub.i32 69 %tmp1 = load <4 x i32>, <4 x i32>* %A 70 %tmp2 = load <4 x i32>, <4 x i32>* %B 71 %tmp3 = sub <4 x i32> %tmp1, %tmp2 72 ret <4 x i32> %tmp3 73} 74 75define <2 x i64> @vsubQi64(<2 x i64>* %A, <2 x i64>* %B) nounwind { 76;CHECK-LABEL: vsubQi64: 77;CHECK: vsub.i64 78 %tmp1 = load <2 x i64>, <2 x i64>* %A 79 %tmp2 = load <2 x i64>, <2 x i64>* %B 80 %tmp3 = sub <2 x i64> %tmp1, %tmp2 81 ret <2 x i64> %tmp3 82} 83 84define <4 x float> @vsubQf32(<4 x float>* %A, <4 x float>* %B) nounwind { 85;CHECK-LABEL: vsubQf32: 86;CHECK: vsub.f32 87 %tmp1 = load <4 x float>, <4 x float>* %A 88 %tmp2 = load <4 x float>, <4 x float>* %B 89 %tmp3 = fsub <4 x float> %tmp1, %tmp2 90 ret <4 x float> %tmp3 91} 92 93define <8 x i8> @vsubhni16_natural(<8 x i16> %A, <8 x i16> %B) nounwind { 94; CHECK-LABEL: vsubhni16_natural: 95; CHECK: vsubhn.i16 96 %sum = sub <8 x i16> %A, %B 97 %shift = lshr <8 x i16> %sum, <i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8, i16 8> 98 %trunc = trunc <8 x i16> %shift to <8 x i8> 99 ret <8 x i8> %trunc 100} 101 102define <4 x i16> @vsubhni32_natural(<4 x i32> %A, <4 x i32> %B) nounwind { 103; CHECK-LABEL: vsubhni32_natural: 104; CHECK: vsubhn.i32 105 %sum = sub <4 x i32> %A, %B 106 %shift = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16> 107 %trunc = trunc <4 x i32> %shift to <4 x i16> 108 ret <4 x i16> %trunc 109} 110 111define <2 x i32> @vsubhni64_natural(<2 x i64> %A, <2 x i64> %B) nounwind { 112; CHECK-LABEL: vsubhni64_natural: 113; CHECK: vsubhn.i64 114 %sum = sub <2 x i64> %A, %B 115 %shift = lshr <2 x i64> %sum, <i64 32, i64 32> 116 %trunc = trunc <2 x i64> %shift to <2 x i32> 117 ret <2 x i32> %trunc 118} 119 120define <8 x i8> @vrsubhni16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 121;CHECK-LABEL: vrsubhni16: 122;CHECK: vrsubhn.i16 123 %tmp1 = load <8 x i16>, <8 x i16>* %A 124 %tmp2 = load <8 x i16>, <8 x i16>* %B 125 %tmp3 = call <8 x i8> @llvm.arm.neon.vrsubhn.v8i8(<8 x i16> %tmp1, <8 x i16> %tmp2) 126 ret <8 x i8> %tmp3 127} 128 129define <4 x i16> @vrsubhni32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 130;CHECK-LABEL: vrsubhni32: 131;CHECK: vrsubhn.i32 132 %tmp1 = load <4 x i32>, <4 x i32>* %A 133 %tmp2 = load <4 x i32>, <4 x i32>* %B 134 %tmp3 = call <4 x i16> @llvm.arm.neon.vrsubhn.v4i16(<4 x i32> %tmp1, <4 x i32> %tmp2) 135 ret <4 x i16> %tmp3 136} 137 138define <2 x i32> @vrsubhni64(<2 x i64>* %A, <2 x i64>* %B) nounwind { 139;CHECK-LABEL: vrsubhni64: 140;CHECK: vrsubhn.i64 141 %tmp1 = load <2 x i64>, <2 x i64>* %A 142 %tmp2 = load <2 x i64>, <2 x i64>* %B 143 %tmp3 = call <2 x i32> @llvm.arm.neon.vrsubhn.v2i32(<2 x i64> %tmp1, <2 x i64> %tmp2) 144 ret <2 x i32> %tmp3 145} 146 147declare <8 x i8> @llvm.arm.neon.vrsubhn.v8i8(<8 x i16>, <8 x i16>) nounwind readnone 148declare <4 x i16> @llvm.arm.neon.vrsubhn.v4i16(<4 x i32>, <4 x i32>) nounwind readnone 149declare <2 x i32> @llvm.arm.neon.vrsubhn.v2i32(<2 x i64>, <2 x i64>) nounwind readnone 150 151define <8 x i16> @vsubls8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 152;CHECK-LABEL: vsubls8: 153;CHECK: vsubl.s8 154 %tmp1 = load <8 x i8>, <8 x i8>* %A 155 %tmp2 = load <8 x i8>, <8 x i8>* %B 156 %tmp3 = sext <8 x i8> %tmp1 to <8 x i16> 157 %tmp4 = sext <8 x i8> %tmp2 to <8 x i16> 158 %tmp5 = sub <8 x i16> %tmp3, %tmp4 159 ret <8 x i16> %tmp5 160} 161 162define <4 x i32> @vsubls16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 163;CHECK-LABEL: vsubls16: 164;CHECK: vsubl.s16 165 %tmp1 = load <4 x i16>, <4 x i16>* %A 166 %tmp2 = load <4 x i16>, <4 x i16>* %B 167 %tmp3 = sext <4 x i16> %tmp1 to <4 x i32> 168 %tmp4 = sext <4 x i16> %tmp2 to <4 x i32> 169 %tmp5 = sub <4 x i32> %tmp3, %tmp4 170 ret <4 x i32> %tmp5 171} 172 173define <2 x i64> @vsubls32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 174;CHECK-LABEL: vsubls32: 175;CHECK: vsubl.s32 176 %tmp1 = load <2 x i32>, <2 x i32>* %A 177 %tmp2 = load <2 x i32>, <2 x i32>* %B 178 %tmp3 = sext <2 x i32> %tmp1 to <2 x i64> 179 %tmp4 = sext <2 x i32> %tmp2 to <2 x i64> 180 %tmp5 = sub <2 x i64> %tmp3, %tmp4 181 ret <2 x i64> %tmp5 182} 183 184define <8 x i16> @vsublu8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 185;CHECK-LABEL: vsublu8: 186;CHECK: vsubl.u8 187 %tmp1 = load <8 x i8>, <8 x i8>* %A 188 %tmp2 = load <8 x i8>, <8 x i8>* %B 189 %tmp3 = zext <8 x i8> %tmp1 to <8 x i16> 190 %tmp4 = zext <8 x i8> %tmp2 to <8 x i16> 191 %tmp5 = sub <8 x i16> %tmp3, %tmp4 192 ret <8 x i16> %tmp5 193} 194 195define <4 x i32> @vsublu16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 196;CHECK-LABEL: vsublu16: 197;CHECK: vsubl.u16 198 %tmp1 = load <4 x i16>, <4 x i16>* %A 199 %tmp2 = load <4 x i16>, <4 x i16>* %B 200 %tmp3 = zext <4 x i16> %tmp1 to <4 x i32> 201 %tmp4 = zext <4 x i16> %tmp2 to <4 x i32> 202 %tmp5 = sub <4 x i32> %tmp3, %tmp4 203 ret <4 x i32> %tmp5 204} 205 206define <2 x i64> @vsublu32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 207;CHECK-LABEL: vsublu32: 208;CHECK: vsubl.u32 209 %tmp1 = load <2 x i32>, <2 x i32>* %A 210 %tmp2 = load <2 x i32>, <2 x i32>* %B 211 %tmp3 = zext <2 x i32> %tmp1 to <2 x i64> 212 %tmp4 = zext <2 x i32> %tmp2 to <2 x i64> 213 %tmp5 = sub <2 x i64> %tmp3, %tmp4 214 ret <2 x i64> %tmp5 215} 216 217define <8 x i16> @vsubws8(<8 x i16>* %A, <8 x i8>* %B) nounwind { 218;CHECK-LABEL: vsubws8: 219;CHECK: vsubw.s8 220 %tmp1 = load <8 x i16>, <8 x i16>* %A 221 %tmp2 = load <8 x i8>, <8 x i8>* %B 222 %tmp3 = sext <8 x i8> %tmp2 to <8 x i16> 223 %tmp4 = sub <8 x i16> %tmp1, %tmp3 224 ret <8 x i16> %tmp4 225} 226 227define <4 x i32> @vsubws16(<4 x i32>* %A, <4 x i16>* %B) nounwind { 228;CHECK-LABEL: vsubws16: 229;CHECK: vsubw.s16 230 %tmp1 = load <4 x i32>, <4 x i32>* %A 231 %tmp2 = load <4 x i16>, <4 x i16>* %B 232 %tmp3 = sext <4 x i16> %tmp2 to <4 x i32> 233 %tmp4 = sub <4 x i32> %tmp1, %tmp3 234 ret <4 x i32> %tmp4 235} 236 237define <2 x i64> @vsubws32(<2 x i64>* %A, <2 x i32>* %B) nounwind { 238;CHECK-LABEL: vsubws32: 239;CHECK: vsubw.s32 240 %tmp1 = load <2 x i64>, <2 x i64>* %A 241 %tmp2 = load <2 x i32>, <2 x i32>* %B 242 %tmp3 = sext <2 x i32> %tmp2 to <2 x i64> 243 %tmp4 = sub <2 x i64> %tmp1, %tmp3 244 ret <2 x i64> %tmp4 245} 246 247define <8 x i16> @vsubwu8(<8 x i16>* %A, <8 x i8>* %B) nounwind { 248;CHECK-LABEL: vsubwu8: 249;CHECK: vsubw.u8 250 %tmp1 = load <8 x i16>, <8 x i16>* %A 251 %tmp2 = load <8 x i8>, <8 x i8>* %B 252 %tmp3 = zext <8 x i8> %tmp2 to <8 x i16> 253 %tmp4 = sub <8 x i16> %tmp1, %tmp3 254 ret <8 x i16> %tmp4 255} 256 257define <4 x i32> @vsubwu16(<4 x i32>* %A, <4 x i16>* %B) nounwind { 258;CHECK-LABEL: vsubwu16: 259;CHECK: vsubw.u16 260 %tmp1 = load <4 x i32>, <4 x i32>* %A 261 %tmp2 = load <4 x i16>, <4 x i16>* %B 262 %tmp3 = zext <4 x i16> %tmp2 to <4 x i32> 263 %tmp4 = sub <4 x i32> %tmp1, %tmp3 264 ret <4 x i32> %tmp4 265} 266 267define <2 x i64> @vsubwu32(<2 x i64>* %A, <2 x i32>* %B) nounwind { 268;CHECK-LABEL: vsubwu32: 269;CHECK: vsubw.u32 270 %tmp1 = load <2 x i64>, <2 x i64>* %A 271 %tmp2 = load <2 x i32>, <2 x i32>* %B 272 %tmp3 = zext <2 x i32> %tmp2 to <2 x i64> 273 %tmp4 = sub <2 x i64> %tmp1, %tmp3 274 ret <2 x i64> %tmp4 275} 276