1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -mtriple=arm64-apple-iphoneos -o - %s | FileCheck %s 3 4define i3 @sign_i3(i3 %a) { 5; CHECK-LABEL: sign_i3: 6; CHECK: // %bb.0: 7; CHECK-NEXT: sbfx w8, w0, #2, #1 8; CHECK-NEXT: orr w0, w8, #0x1 9; CHECK-NEXT: ret 10 %c = icmp sgt i3 %a, -1 11 %res = select i1 %c, i3 1, i3 -1 12 ret i3 %res 13} 14 15define i4 @sign_i4(i4 %a) { 16; CHECK-LABEL: sign_i4: 17; CHECK: // %bb.0: 18; CHECK-NEXT: sbfx w8, w0, #3, #1 19; CHECK-NEXT: orr w0, w8, #0x1 20; CHECK-NEXT: ret 21 %c = icmp sgt i4 %a, -1 22 %res = select i1 %c, i4 1, i4 -1 23 ret i4 %res 24} 25 26define i8 @sign_i8(i8 %a) { 27; CHECK-LABEL: sign_i8: 28; CHECK: // %bb.0: 29; CHECK-NEXT: sbfx w8, w0, #7, #1 30; CHECK-NEXT: orr w0, w8, #0x1 31; CHECK-NEXT: ret 32 %c = icmp sgt i8 %a, -1 33 %res = select i1 %c, i8 1, i8 -1 34 ret i8 %res 35} 36 37define i16 @sign_i16(i16 %a) { 38; CHECK-LABEL: sign_i16: 39; CHECK: // %bb.0: 40; CHECK-NEXT: sbfx w8, w0, #15, #1 41; CHECK-NEXT: orr w0, w8, #0x1 42; CHECK-NEXT: ret 43 %c = icmp sgt i16 %a, -1 44 %res = select i1 %c, i16 1, i16 -1 45 ret i16 %res 46} 47 48define i32 @sign_i32(i32 %a) { 49; CHECK-LABEL: sign_i32: 50; CHECK: // %bb.0: 51; CHECK-NEXT: asr w8, w0, #31 52; CHECK-NEXT: orr w0, w8, #0x1 53; CHECK-NEXT: ret 54 %c = icmp sgt i32 %a, -1 55 %res = select i1 %c, i32 1, i32 -1 56 ret i32 %res 57} 58 59define i64 @sign_i64(i64 %a) { 60; CHECK-LABEL: sign_i64: 61; CHECK: // %bb.0: 62; CHECK-NEXT: asr x8, x0, #63 63; CHECK-NEXT: orr x0, x8, #0x1 64; CHECK-NEXT: ret 65 %c = icmp sgt i64 %a, -1 66 %res = select i1 %c, i64 1, i64 -1 67 ret i64 %res 68} 69 70 71define i64 @not_sign_i64(i64 %a) { 72; CHECK-LABEL: not_sign_i64: 73; CHECK: // %bb.0: 74; CHECK-NEXT: cmp x0, #0 // =0 75; CHECK-NEXT: mov w8, #1 76; CHECK-NEXT: cneg x0, x8, le 77; CHECK-NEXT: ret 78 %c = icmp sgt i64 %a, 0 79 %res = select i1 %c, i64 1, i64 -1 80 ret i64 %res 81} 82 83define i64 @not_sign_i64_2(i64 %a) { 84; CHECK-LABEL: not_sign_i64_2: 85; CHECK: // %bb.0: 86; CHECK-NEXT: asr x0, x0, #63 87; CHECK-NEXT: ret 88 %c = icmp sgt i64 %a, -1 89 %res = select i1 %c, i64 0, i64 -1 90 ret i64 %res 91} 92 93define i64 @not_sign_i64_3(i64 %a) { 94; CHECK-LABEL: not_sign_i64_3: 95; CHECK: // %bb.0: 96; CHECK-NEXT: mvn x8, x0 97; CHECK-NEXT: lsr x0, x8, #63 98; CHECK-NEXT: ret 99 %c = icmp sgt i64 %a, -1 100 %res = select i1 %c, i64 1, i64 0 101 ret i64 %res 102} 103 104define i64 @not_sign_i64_4(i64 %a) { 105; CHECK-LABEL: not_sign_i64_4: 106; CHECK: // %bb.0: 107; CHECK-NEXT: mov x0, #-1 108; CHECK-NEXT: ret 109 %c = icmp ugt i64 %a, -1 110 %res = select i1 %c, i64 1, i64 -1 111 ret i64 %res 112} 113 114define <7 x i8> @sign_7xi8(<7 x i8> %a) { 115; CHECK-LABEL: sign_7xi8: 116; CHECK: // %bb.0: 117; CHECK-NEXT: sshr v0.8b, v0.8b, #7 118; CHECK-NEXT: movi v1.8b, #1 119; CHECK-NEXT: orr v0.8b, v0.8b, v1.8b 120; CHECK-NEXT: ret 121 %c = icmp sgt <7 x i8> %a, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1> 122 %res = select <7 x i1> %c, <7 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, <7 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1> 123 ret <7 x i8> %res 124} 125 126define <8 x i8> @sign_8xi8(<8 x i8> %a) { 127; CHECK-LABEL: sign_8xi8: 128; CHECK: // %bb.0: 129; CHECK-NEXT: sshr v0.8b, v0.8b, #7 130; CHECK-NEXT: movi v1.8b, #1 131; CHECK-NEXT: orr v0.8b, v0.8b, v1.8b 132; CHECK-NEXT: ret 133 %c = icmp sgt <8 x i8> %a, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1> 134 %res = select <8 x i1> %c, <8 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, <8 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1> 135 ret <8 x i8> %res 136} 137 138define <16 x i8> @sign_16xi8(<16 x i8> %a) { 139; CHECK-LABEL: sign_16xi8: 140; CHECK: // %bb.0: 141; CHECK-NEXT: sshr v0.16b, v0.16b, #7 142; CHECK-NEXT: movi v1.16b, #1 143; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b 144; CHECK-NEXT: ret 145 %c = icmp sgt <16 x i8> %a, <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1> 146 %res = select <16 x i1> %c, <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1> 147 ret <16 x i8> %res 148} 149 150define <3 x i32> @sign_3xi32(<3 x i32> %a) { 151; CHECK-LABEL: sign_3xi32: 152; CHECK: // %bb.0: 153; CHECK-NEXT: sshr v0.4s, v0.4s, #31 154; CHECK-NEXT: orr v0.4s, #1 155; CHECK-NEXT: ret 156 %c = icmp sgt <3 x i32> %a, <i32 -1, i32 -1, i32 -1> 157 %res = select <3 x i1> %c, <3 x i32> <i32 1, i32 1, i32 1>, <3 x i32> <i32 -1, i32 -1, i32 -1> 158 ret <3 x i32> %res 159} 160 161define <4 x i32> @sign_4xi32(<4 x i32> %a) { 162; CHECK-LABEL: sign_4xi32: 163; CHECK: // %bb.0: 164; CHECK-NEXT: sshr v0.4s, v0.4s, #31 165; CHECK-NEXT: orr v0.4s, #1 166; CHECK-NEXT: ret 167 %c = icmp sgt <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1> 168 %res = select <4 x i1> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1> 169 ret <4 x i32> %res 170} 171 172define <4 x i32> @sign_4xi32_multi_use(<4 x i32> %a) { 173; CHECK-LABEL: sign_4xi32_multi_use: 174; CHECK: // %bb.0: 175; CHECK-NEXT: sub sp, sp, #32 // =32 176; CHECK-NEXT: str x30, [sp, #16] // 8-byte Folded Spill 177; CHECK-NEXT: .cfi_def_cfa_offset 32 178; CHECK-NEXT: .cfi_offset w30, -16 179; CHECK-NEXT: movi v1.2d, #0xffffffffffffffff 180; CHECK-NEXT: sshr v2.4s, v0.4s, #31 181; CHECK-NEXT: cmgt v0.4s, v0.4s, v1.4s 182; CHECK-NEXT: orr v2.4s, #1 183; CHECK-NEXT: xtn v0.4h, v0.4s 184; CHECK-NEXT: str q2, [sp] // 16-byte Folded Spill 185; CHECK-NEXT: bl use_4xi1 186; CHECK-NEXT: ldr q0, [sp] // 16-byte Folded Reload 187; CHECK-NEXT: ldr x30, [sp, #16] // 8-byte Folded Reload 188; CHECK-NEXT: add sp, sp, #32 // =32 189; CHECK-NEXT: ret 190 %c = icmp sgt <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1> 191 %res = select <4 x i1> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1> 192 call void @use_4xi1(<4 x i1> %c) 193 ret <4 x i32> %res 194} 195 196; Second icmp operand breaks sign pattern. 197define <4 x i32> @not_sign_4xi32(<4 x i32> %a) { 198; CHECK-LABEL: not_sign_4xi32: 199; CHECK: // %bb.0: 200; CHECK-NEXT: adrp x8, .LCPI16_0 201; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI16_0] 202; CHECK-NEXT: cmgt v0.4s, v0.4s, v1.4s 203; CHECK-NEXT: movi v1.4s, #1 204; CHECK-NEXT: and v1.16b, v0.16b, v1.16b 205; CHECK-NEXT: orn v0.16b, v1.16b, v0.16b 206; CHECK-NEXT: ret 207 %c = icmp sgt <4 x i32> %a, <i32 1, i32 -1, i32 -1, i32 -1> 208 %res = select <4 x i1> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1> 209 ret <4 x i32> %res 210} 211 212; First select operand breaks sign pattern. 213define <4 x i32> @not_sign_4xi32_2(<4 x i32> %a) { 214; CHECK-LABEL: not_sign_4xi32_2: 215; CHECK: // %bb.0: 216; CHECK-NEXT: adrp x8, .LCPI17_0 217; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI17_0] 218; CHECK-NEXT: movi v2.2d, #0xffffffffffffffff 219; CHECK-NEXT: cmgt v0.4s, v0.4s, v2.4s 220; CHECK-NEXT: and v1.16b, v0.16b, v1.16b 221; CHECK-NEXT: orn v0.16b, v1.16b, v0.16b 222; CHECK-NEXT: ret 223 %c = icmp sgt <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1> 224 %res = select <4 x i1> %c, <4 x i32> <i32 1, i32 1, i32 -1, i32 1>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1> 225 ret <4 x i32> %res 226} 227 228; Second select operand breaks sign pattern. 229define <4 x i32> @not_sign_4xi32_3(<4 x i32> %a) { 230; CHECK-LABEL: not_sign_4xi32_3: 231; CHECK: // %bb.0: 232; CHECK-NEXT: adrp x8, .LCPI18_0 233; CHECK-NEXT: ldr q1, [x8, :lo12:.LCPI18_0] 234; CHECK-NEXT: movi v2.2d, #0xffffffffffffffff 235; CHECK-NEXT: cmgt v0.4s, v0.4s, v2.4s 236; CHECK-NEXT: movi v2.4s, #1 237; CHECK-NEXT: bsl v0.16b, v2.16b, v1.16b 238; CHECK-NEXT: ret 239 %c = icmp sgt <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1> 240 %res = select <4 x i1> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 1> 241 ret <4 x i32> %res 242} 243 244; i65 is not legal. 245define <4 x i65> @sign_4xi65(<4 x i65> %a) { 246; CHECK-LABEL: sign_4xi65: 247; CHECK: // %bb.0: 248; CHECK-NEXT: sbfx x8, x1, #0, #1 249; CHECK-NEXT: sbfx x9, x7, #0, #1 250; CHECK-NEXT: orr x6, x9, #0x1 251; CHECK-NEXT: lsr x7, x9, #63 252; CHECK-NEXT: orr x9, x8, #0x1 253; CHECK-NEXT: lsr x1, x8, #63 254; CHECK-NEXT: fmov d0, x9 255; CHECK-NEXT: sbfx x10, x5, #0, #1 256; CHECK-NEXT: sbfx x11, x3, #0, #1 257; CHECK-NEXT: mov v0.d[1], x1 258; CHECK-NEXT: orr x2, x11, #0x1 259; CHECK-NEXT: lsr x3, x11, #63 260; CHECK-NEXT: orr x4, x10, #0x1 261; CHECK-NEXT: lsr x5, x10, #63 262; CHECK-NEXT: fmov x0, d0 263; CHECK-NEXT: ret 264 %c = icmp sgt <4 x i65> %a, <i65 -1, i65 -1, i65 -1, i65 -1> 265 %res = select <4 x i1> %c, <4 x i65> <i65 1, i65 1, i65 1, i65 1>, <4 x i65 > <i65 -1, i65 -1, i65 -1, i65 -1> 266 ret <4 x i65> %res 267} 268 269declare void @use_4xi1(<4 x i1>) 270