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; INSERT VECTOR ELT 5 6define <vscale x 8 x i8> @promote_insert_8i8(<vscale x 8 x i8> %a, i8 %elt, i64 %idx) { 7; CHECK-LABEL: promote_insert_8i8: 8; CHECK: // %bb.0: 9; CHECK-NEXT: index z1.h, #0, #1 10; CHECK-NEXT: mov z2.h, w1 11; CHECK-NEXT: ptrue p0.h 12; CHECK-NEXT: cmpeq p0.h, p0/z, z1.h, z2.h 13; CHECK-NEXT: mov z0.h, p0/m, w0 14; CHECK-NEXT: ret 15 %ins = insertelement <vscale x 8 x i8> %a, i8 %elt, i64 %idx 16 ret <vscale x 8 x i8> %ins 17} 18 19define <vscale x 32 x i8> @split_insert_32i8_idx(<vscale x 32 x i8> %a, i8 %elt, i64 %idx) { 20; CHECK-LABEL: split_insert_32i8_idx: 21; CHECK: // %bb.0: 22; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 23; CHECK-NEXT: addvl sp, sp, #-2 24; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 25; CHECK-NEXT: .cfi_offset w29, -16 26; CHECK-NEXT: rdvl x8, #2 27; CHECK-NEXT: ptrue p0.b 28; CHECK-NEXT: mov x9, sp 29; CHECK-NEXT: sub x8, x8, #1 30; CHECK-NEXT: cmp x1, x8 31; CHECK-NEXT: st1b { z1.b }, p0, [sp, #1, mul vl] 32; CHECK-NEXT: csel x8, x1, x8, lo 33; CHECK-NEXT: st1b { z0.b }, p0, [sp] 34; CHECK-NEXT: strb w0, [x9, x8] 35; CHECK-NEXT: ld1b { z0.b }, p0/z, [sp] 36; CHECK-NEXT: ld1b { z1.b }, p0/z, [sp, #1, mul vl] 37; CHECK-NEXT: addvl sp, sp, #2 38; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 39; CHECK-NEXT: ret 40 %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 %idx 41 ret <vscale x 32 x i8> %ins 42} 43 44define <vscale x 8 x float> @split_insert_8f32_idx(<vscale x 8 x float> %a, float %elt, i64 %idx) { 45; CHECK-LABEL: split_insert_8f32_idx: 46; CHECK: // %bb.0: 47; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 48; CHECK-NEXT: addvl sp, sp, #-2 49; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 50; CHECK-NEXT: .cfi_offset w29, -16 51; CHECK-NEXT: cnth x8 52; CHECK-NEXT: ptrue p0.s 53; CHECK-NEXT: mov x9, sp 54; CHECK-NEXT: sub x8, x8, #1 55; CHECK-NEXT: cmp x0, x8 56; CHECK-NEXT: st1w { z1.s }, p0, [sp, #1, mul vl] 57; CHECK-NEXT: csel x8, x0, x8, lo 58; CHECK-NEXT: st1w { z0.s }, p0, [sp] 59; CHECK-NEXT: str s2, [x9, x8, lsl #2] 60; CHECK-NEXT: ld1w { z0.s }, p0/z, [sp] 61; CHECK-NEXT: ld1w { z1.s }, p0/z, [sp, #1, mul vl] 62; CHECK-NEXT: addvl sp, sp, #2 63; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 64; CHECK-NEXT: ret 65 %ins = insertelement <vscale x 8 x float> %a, float %elt, i64 %idx 66 ret <vscale x 8 x float> %ins 67} 68 69define <vscale x 8 x i64> @split_insert_8i64_idx(<vscale x 8 x i64> %a, i64 %elt, i64 %idx) { 70; CHECK-LABEL: split_insert_8i64_idx: 71; CHECK: // %bb.0: 72; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 73; CHECK-NEXT: addvl sp, sp, #-4 74; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG 75; CHECK-NEXT: .cfi_offset w29, -16 76; CHECK-NEXT: cnth x8 77; CHECK-NEXT: ptrue p0.d 78; CHECK-NEXT: mov x9, sp 79; CHECK-NEXT: sub x8, x8, #1 80; CHECK-NEXT: cmp x1, x8 81; CHECK-NEXT: st1d { z3.d }, p0, [sp, #3, mul vl] 82; CHECK-NEXT: csel x8, x1, x8, lo 83; CHECK-NEXT: st1d { z2.d }, p0, [sp, #2, mul vl] 84; CHECK-NEXT: st1d { z1.d }, p0, [sp, #1, mul vl] 85; CHECK-NEXT: st1d { z0.d }, p0, [sp] 86; CHECK-NEXT: str x0, [x9, x8, lsl #3] 87; CHECK-NEXT: ld1d { z0.d }, p0/z, [sp] 88; CHECK-NEXT: ld1d { z1.d }, p0/z, [sp, #1, mul vl] 89; CHECK-NEXT: ld1d { z2.d }, p0/z, [sp, #2, mul vl] 90; CHECK-NEXT: ld1d { z3.d }, p0/z, [sp, #3, mul vl] 91; CHECK-NEXT: addvl sp, sp, #4 92; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 93; CHECK-NEXT: ret 94 %ins = insertelement <vscale x 8 x i64> %a, i64 %elt, i64 %idx 95 ret <vscale x 8 x i64> %ins 96} 97 98; INSERT VECTOR ELT, CONSTANT IDX 99 100define <vscale x 4 x i16> @promote_insert_4i16(<vscale x 4 x i16> %a, i16 %elt) { 101; CHECK-LABEL: promote_insert_4i16: 102; CHECK: // %bb.0: 103; CHECK-NEXT: mov w8, #5 // =0x5 104; CHECK-NEXT: index z1.s, #0, #1 105; CHECK-NEXT: ptrue p0.s 106; CHECK-NEXT: mov z2.s, w8 107; CHECK-NEXT: cmpeq p0.s, p0/z, z1.s, z2.s 108; CHECK-NEXT: mov z0.s, p0/m, w0 109; CHECK-NEXT: ret 110 %ins = insertelement <vscale x 4 x i16> %a, i16 %elt, i64 5 111 ret <vscale x 4 x i16> %ins 112} 113 114; In this test, the index is small enough that we know it will be in the 115; low half of the vector and there is no need to go through the stack as 116; done in the remaining tests 117define <vscale x 32 x i8> @split_insert_32i8(<vscale x 32 x i8> %a, i8 %elt) { 118; CHECK-LABEL: split_insert_32i8: 119; CHECK: // %bb.0: 120; CHECK-NEXT: mov w8, #3 // =0x3 121; CHECK-NEXT: index z2.b, #0, #1 122; CHECK-NEXT: ptrue p0.b 123; CHECK-NEXT: mov z3.b, w8 124; CHECK-NEXT: cmpeq p0.b, p0/z, z2.b, z3.b 125; CHECK-NEXT: mov z0.b, p0/m, w0 126; CHECK-NEXT: ret 127 %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 3 128 ret <vscale x 32 x i8> %ins 129} 130 131define <vscale x 32 x i16> @split_insert_32i16(<vscale x 32 x i16> %a, i16 %elt) { 132; CHECK-LABEL: split_insert_32i16: 133; CHECK: // %bb.0: 134; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 135; CHECK-NEXT: addvl sp, sp, #-4 136; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG 137; CHECK-NEXT: .cfi_offset w29, -16 138; CHECK-NEXT: rdvl x8, #2 139; CHECK-NEXT: ptrue p0.h 140; CHECK-NEXT: mov w9, #128 // =0x80 141; CHECK-NEXT: sub x8, x8, #1 142; CHECK-NEXT: cmp x8, #128 143; CHECK-NEXT: st1h { z3.h }, p0, [sp, #3, mul vl] 144; CHECK-NEXT: csel x8, x8, x9, lo 145; CHECK-NEXT: mov x9, sp 146; CHECK-NEXT: st1h { z2.h }, p0, [sp, #2, mul vl] 147; CHECK-NEXT: st1h { z1.h }, p0, [sp, #1, mul vl] 148; CHECK-NEXT: st1h { z0.h }, p0, [sp] 149; CHECK-NEXT: strh w0, [x9, x8, lsl #1] 150; CHECK-NEXT: ld1h { z0.h }, p0/z, [sp] 151; CHECK-NEXT: ld1h { z1.h }, p0/z, [sp, #1, mul vl] 152; CHECK-NEXT: ld1h { z2.h }, p0/z, [sp, #2, mul vl] 153; CHECK-NEXT: ld1h { z3.h }, p0/z, [sp, #3, mul vl] 154; CHECK-NEXT: addvl sp, sp, #4 155; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 156; CHECK-NEXT: ret 157 %ins = insertelement <vscale x 32 x i16> %a, i16 %elt, i64 128 158 ret <vscale x 32 x i16> %ins 159} 160 161define <vscale x 8 x i32> @split_insert_8i32(<vscale x 8 x i32> %a, i32 %elt) { 162; CHECK-LABEL: split_insert_8i32: 163; CHECK: // %bb.0: 164; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 165; CHECK-NEXT: addvl sp, sp, #-2 166; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 167; CHECK-NEXT: .cfi_offset w29, -16 168; CHECK-NEXT: cnth x8 169; CHECK-NEXT: mov w9, #16960 // =0x4240 170; CHECK-NEXT: ptrue p0.s 171; CHECK-NEXT: movk w9, #15, lsl #16 172; CHECK-NEXT: sub x8, x8, #1 173; CHECK-NEXT: cmp x8, x9 174; CHECK-NEXT: st1w { z1.s }, p0, [sp, #1, mul vl] 175; CHECK-NEXT: csel x8, x8, x9, lo 176; CHECK-NEXT: mov x9, sp 177; CHECK-NEXT: st1w { z0.s }, p0, [sp] 178; CHECK-NEXT: str w0, [x9, x8, lsl #2] 179; CHECK-NEXT: ld1w { z0.s }, p0/z, [sp] 180; CHECK-NEXT: ld1w { z1.s }, p0/z, [sp, #1, mul vl] 181; CHECK-NEXT: addvl sp, sp, #2 182; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 183; CHECK-NEXT: ret 184 %ins = insertelement <vscale x 8 x i32> %a, i32 %elt, i64 1000000 185 ret <vscale x 8 x i32> %ins 186} 187