1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s 3 4; EXTRACT VECTOR ELT 5 6define i32 @promote_extract_2i32_idx(<vscale x 2 x i32> %a, i32 %idx) { 7; CHECK-LABEL: promote_extract_2i32_idx: 8; CHECK: // %bb.0: 9; CHECK-NEXT: mov w8, w0 10; CHECK-NEXT: whilels p0.d, xzr, x8 11; CHECK-NEXT: lastb x0, p0, z0.d 12; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0 13; CHECK-NEXT: ret 14 %ext = extractelement <vscale x 2 x i32> %a, i32 %idx 15 ret i32 %ext 16} 17 18define i8 @split_extract_32i8_idx(<vscale x 32 x i8> %a, i32 %idx) { 19; CHECK-LABEL: split_extract_32i8_idx: 20; CHECK: // %bb.0: 21; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 22; CHECK-NEXT: addvl sp, sp, #-2 23; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 24; CHECK-NEXT: .cfi_offset w29, -16 25; CHECK-NEXT: rdvl x8, #2 26; CHECK-NEXT: ptrue p0.b 27; CHECK-NEXT: mov w9, w0 28; CHECK-NEXT: sub x8, x8, #1 29; CHECK-NEXT: cmp x9, x8 30; CHECK-NEXT: st1b { z1.b }, p0, [sp, #1, mul vl] 31; CHECK-NEXT: st1b { z0.b }, p0, [sp] 32; CHECK-NEXT: csel x8, x9, x8, lo 33; CHECK-NEXT: mov x9, sp 34; CHECK-NEXT: ldrb w0, [x9, x8] 35; CHECK-NEXT: addvl sp, sp, #2 36; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 37; CHECK-NEXT: ret 38 %ext = extractelement <vscale x 32 x i8> %a, i32 %idx 39 ret i8 %ext 40} 41 42define i16 @split_extract_16i16_idx(<vscale x 16 x i16> %a, i32 %idx) { 43; CHECK-LABEL: split_extract_16i16_idx: 44; CHECK: // %bb.0: 45; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 46; CHECK-NEXT: addvl sp, sp, #-2 47; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 48; CHECK-NEXT: .cfi_offset w29, -16 49; CHECK-NEXT: rdvl x8, #1 50; CHECK-NEXT: ptrue p0.h 51; CHECK-NEXT: mov w9, w0 52; CHECK-NEXT: sub x8, x8, #1 53; CHECK-NEXT: cmp x9, x8 54; CHECK-NEXT: st1h { z1.h }, p0, [sp, #1, mul vl] 55; CHECK-NEXT: st1h { z0.h }, p0, [sp] 56; CHECK-NEXT: csel x8, x9, x8, lo 57; CHECK-NEXT: mov x9, sp 58; CHECK-NEXT: ldrh w0, [x9, x8, lsl #1] 59; CHECK-NEXT: addvl sp, sp, #2 60; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 61; CHECK-NEXT: ret 62 %ext = extractelement <vscale x 16 x i16> %a, i32 %idx 63 ret i16 %ext 64} 65 66define i32 @split_extract_8i32_idx(<vscale x 8 x i32> %a, i32 %idx) { 67; CHECK-LABEL: split_extract_8i32_idx: 68; CHECK: // %bb.0: 69; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 70; CHECK-NEXT: addvl sp, sp, #-2 71; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 72; CHECK-NEXT: .cfi_offset w29, -16 73; CHECK-NEXT: cnth x8 74; CHECK-NEXT: ptrue p0.s 75; CHECK-NEXT: mov w9, w0 76; CHECK-NEXT: sub x8, x8, #1 77; CHECK-NEXT: cmp x9, x8 78; CHECK-NEXT: st1w { z1.s }, p0, [sp, #1, mul vl] 79; CHECK-NEXT: st1w { z0.s }, p0, [sp] 80; CHECK-NEXT: csel x8, x9, x8, lo 81; CHECK-NEXT: mov x9, sp 82; CHECK-NEXT: ldr w0, [x9, x8, lsl #2] 83; CHECK-NEXT: addvl sp, sp, #2 84; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 85; CHECK-NEXT: ret 86 %ext = extractelement <vscale x 8 x i32> %a, i32 %idx 87 ret i32 %ext 88} 89 90define i64 @split_extract_8i64_idx(<vscale x 8 x i64> %a, i32 %idx) { 91; CHECK-LABEL: split_extract_8i64_idx: 92; CHECK: // %bb.0: 93; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 94; CHECK-NEXT: addvl sp, sp, #-4 95; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG 96; CHECK-NEXT: .cfi_offset w29, -16 97; CHECK-NEXT: cnth x8 98; CHECK-NEXT: ptrue p0.d 99; CHECK-NEXT: mov w9, w0 100; CHECK-NEXT: sub x8, x8, #1 101; CHECK-NEXT: cmp x9, x8 102; CHECK-NEXT: st1d { z3.d }, p0, [sp, #3, mul vl] 103; CHECK-NEXT: st1d { z2.d }, p0, [sp, #2, mul vl] 104; CHECK-NEXT: csel x8, x9, x8, lo 105; CHECK-NEXT: mov x9, sp 106; CHECK-NEXT: st1d { z1.d }, p0, [sp, #1, mul vl] 107; CHECK-NEXT: st1d { z0.d }, p0, [sp] 108; CHECK-NEXT: ldr x0, [x9, x8, lsl #3] 109; CHECK-NEXT: addvl sp, sp, #4 110; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 111; CHECK-NEXT: ret 112 %ext = extractelement <vscale x 8 x i64> %a, i32 %idx 113 ret i64 %ext 114} 115 116; EXTRACT VECTOR ELT, CONSTANT IDX 117 118define i16 @promote_extract_4i16(<vscale x 4 x i16> %a) { 119; CHECK-LABEL: promote_extract_4i16: 120; CHECK: // %bb.0: 121; CHECK-NEXT: mov w0, v0.s[1] 122; CHECK-NEXT: ret 123 %ext = extractelement <vscale x 4 x i16> %a, i32 1 124 ret i16 %ext 125} 126 127define i8 @split_extract_32i8(<vscale x 32 x i8> %a) { 128; CHECK-LABEL: split_extract_32i8: 129; CHECK: // %bb.0: 130; CHECK-NEXT: umov w0, v0.b[3] 131; CHECK-NEXT: ret 132 %ext = extractelement <vscale x 32 x i8> %a, i32 3 133 ret i8 %ext 134} 135 136define i16 @split_extract_16i16(<vscale x 16 x i16> %a) { 137; CHECK-LABEL: split_extract_16i16: 138; CHECK: // %bb.0: 139; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 140; CHECK-NEXT: addvl sp, sp, #-2 141; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 142; CHECK-NEXT: .cfi_offset w29, -16 143; CHECK-NEXT: rdvl x8, #1 144; CHECK-NEXT: ptrue p0.h 145; CHECK-NEXT: mov w9, #128 // =0x80 146; CHECK-NEXT: sub x8, x8, #1 147; CHECK-NEXT: cmp x8, #128 148; CHECK-NEXT: st1h { z1.h }, p0, [sp, #1, mul vl] 149; CHECK-NEXT: st1h { z0.h }, p0, [sp] 150; CHECK-NEXT: csel x8, x8, x9, lo 151; CHECK-NEXT: mov x9, sp 152; CHECK-NEXT: ldrh w0, [x9, x8, lsl #1] 153; CHECK-NEXT: addvl sp, sp, #2 154; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 155; CHECK-NEXT: ret 156 %ext = extractelement <vscale x 16 x i16> %a, i32 128 157 ret i16 %ext 158} 159 160define i32 @split_extract_16i32(<vscale x 16 x i32> %a) { 161; CHECK-LABEL: split_extract_16i32: 162; CHECK: // %bb.0: 163; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 164; CHECK-NEXT: addvl sp, sp, #-4 165; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG 166; CHECK-NEXT: .cfi_offset w29, -16 167; CHECK-NEXT: rdvl x8, #1 168; CHECK-NEXT: mov w9, #34464 // =0x86a0 169; CHECK-NEXT: ptrue p0.s 170; CHECK-NEXT: movk w9, #1, lsl #16 171; CHECK-NEXT: sub x8, x8, #1 172; CHECK-NEXT: cmp x8, x9 173; CHECK-NEXT: st1w { z3.s }, p0, [sp, #3, mul vl] 174; CHECK-NEXT: st1w { z2.s }, p0, [sp, #2, mul vl] 175; CHECK-NEXT: csel x8, x8, x9, lo 176; CHECK-NEXT: mov x9, sp 177; CHECK-NEXT: st1w { z1.s }, p0, [sp, #1, mul vl] 178; CHECK-NEXT: st1w { z0.s }, p0, [sp] 179; CHECK-NEXT: ldr w0, [x9, x8, lsl #2] 180; CHECK-NEXT: addvl sp, sp, #4 181; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 182; CHECK-NEXT: ret 183 %ext = extractelement <vscale x 16 x i32> %a, i32 100000 184 ret i32 %ext 185} 186 187define i64 @split_extract_4i64(<vscale x 4 x i64> %a) { 188; CHECK-LABEL: split_extract_4i64: 189; CHECK: // %bb.0: 190; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 191; CHECK-NEXT: addvl sp, sp, #-2 192; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 193; CHECK-NEXT: .cfi_offset w29, -16 194; CHECK-NEXT: cntw x8 195; CHECK-NEXT: ptrue p0.d 196; CHECK-NEXT: mov w9, #10 // =0xa 197; CHECK-NEXT: sub x8, x8, #1 198; CHECK-NEXT: cmp x8, #10 199; CHECK-NEXT: st1d { z1.d }, p0, [sp, #1, mul vl] 200; CHECK-NEXT: st1d { z0.d }, p0, [sp] 201; CHECK-NEXT: csel x8, x8, x9, lo 202; CHECK-NEXT: mov x9, sp 203; CHECK-NEXT: ldr x0, [x9, x8, lsl #3] 204; CHECK-NEXT: addvl sp, sp, #2 205; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 206; CHECK-NEXT: ret 207 %ext = extractelement <vscale x 4 x i64> %a, i32 10 208 ret i64 %ext 209} 210