1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s 2>%t | FileCheck %s 3; RUN: FileCheck --check-prefix=WARN --allow-empty %s <%t 4 5; If this check fails please read test/CodeGen/AArch64/README for instructions on how to resolve it. 6; WARN-NOT: warning 7 8; INSERT VECTOR ELT 9 10define <vscale x 8 x i8> @promote_insert_8i8(<vscale x 8 x i8> %a, i8 %elt, i64 %idx) { 11; CHECK-LABEL: promote_insert_8i8: 12; CHECK: // %bb.0: 13; CHECK-NEXT: mov z1.h, w1 14; CHECK-NEXT: index z2.h, #0, #1 15; CHECK-NEXT: ptrue p0.h 16; CHECK-NEXT: cmpeq p0.h, p0/z, z2.h, z1.h 17; CHECK-NEXT: mov z0.h, p0/m, w0 18; CHECK-NEXT: ret 19 %ins = insertelement <vscale x 8 x i8> %a, i8 %elt, i64 %idx 20 ret <vscale x 8 x i8> %ins 21} 22 23define <vscale x 32 x i8> @split_insert_32i8_idx(<vscale x 32 x i8> %a, i8 %elt, i64 %idx) { 24; CHECK-LABEL: split_insert_32i8_idx: 25; CHECK: // %bb.0: 26; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 27; CHECK-NEXT: addvl sp, sp, #-2 28; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 29; CHECK-NEXT: .cfi_offset w29, -16 30; CHECK-NEXT: rdvl x8, #2 31; CHECK-NEXT: sub x8, x8, #1 // =1 32; CHECK-NEXT: cmp x1, x8 33; CHECK-NEXT: ptrue p0.b 34; CHECK-NEXT: csel x8, x1, x8, lo 35; CHECK-NEXT: mov x9, sp 36; CHECK-NEXT: st1b { z1.b }, p0, [x9, #1, mul vl] 37; CHECK-NEXT: st1b { z0.b }, p0, [sp] 38; CHECK-NEXT: strb w0, [x9, x8] 39; CHECK-NEXT: ld1b { z1.b }, p0/z, [x9, #1, mul vl] 40; CHECK-NEXT: ld1b { z0.b }, p0/z, [sp] 41; CHECK-NEXT: addvl sp, sp, #2 42; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 43; CHECK-NEXT: ret 44 %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 %idx 45 ret <vscale x 32 x i8> %ins 46} 47 48define <vscale x 8 x float> @split_insert_8f32_idx(<vscale x 8 x float> %a, float %elt, i64 %idx) { 49; CHECK-LABEL: split_insert_8f32_idx: 50; CHECK: // %bb.0: 51; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 52; CHECK-NEXT: addvl sp, sp, #-2 53; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 54; CHECK-NEXT: .cfi_offset w29, -16 55; CHECK-NEXT: cnth x8 56; CHECK-NEXT: sub x8, x8, #1 // =1 57; CHECK-NEXT: cmp x0, x8 58; CHECK-NEXT: ptrue p0.s 59; CHECK-NEXT: csel x8, x0, x8, lo 60; CHECK-NEXT: mov x9, sp 61; CHECK-NEXT: st1w { z1.s }, p0, [x9, #1, mul vl] 62; CHECK-NEXT: st1w { z0.s }, p0, [sp] 63; CHECK-NEXT: str s2, [x9, x8, lsl #2] 64; CHECK-NEXT: ld1w { z1.s }, p0/z, [x9, #1, mul vl] 65; CHECK-NEXT: ld1w { z0.s }, p0/z, [sp] 66; CHECK-NEXT: addvl sp, sp, #2 67; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 68; CHECK-NEXT: ret 69 %ins = insertelement <vscale x 8 x float> %a, float %elt, i64 %idx 70 ret <vscale x 8 x float> %ins 71} 72 73define <vscale x 8 x i64> @split_insert_8i64_idx(<vscale x 8 x i64> %a, i64 %elt, i64 %idx) { 74; CHECK-LABEL: split_insert_8i64_idx: 75; CHECK: // %bb.0: 76; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 77; CHECK-NEXT: addvl sp, sp, #-4 78; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG 79; CHECK-NEXT: .cfi_offset w29, -16 80; CHECK-NEXT: cnth x8 81; CHECK-NEXT: sub x8, x8, #1 // =1 82; CHECK-NEXT: cmp x1, x8 83; CHECK-NEXT: ptrue p0.d 84; CHECK-NEXT: csel x8, x1, x8, lo 85; CHECK-NEXT: mov x9, sp 86; CHECK-NEXT: st1d { z3.d }, p0, [x9, #3, mul vl] 87; CHECK-NEXT: st1d { z2.d }, p0, [x9, #2, mul vl] 88; CHECK-NEXT: st1d { z1.d }, p0, [x9, #1, mul vl] 89; CHECK-NEXT: st1d { z0.d }, p0, [sp] 90; CHECK-NEXT: str x0, [x9, x8, lsl #3] 91; CHECK-NEXT: ld1d { z1.d }, p0/z, [x9, #1, mul vl] 92; CHECK-NEXT: ld1d { z2.d }, p0/z, [x9, #2, mul vl] 93; CHECK-NEXT: ld1d { z3.d }, p0/z, [x9, #3, mul vl] 94; CHECK-NEXT: ld1d { z0.d }, p0/z, [sp] 95; CHECK-NEXT: addvl sp, sp, #4 96; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 97; CHECK-NEXT: ret 98 %ins = insertelement <vscale x 8 x i64> %a, i64 %elt, i64 %idx 99 ret <vscale x 8 x i64> %ins 100} 101 102; INSERT VECTOR ELT, CONSTANT IDX 103 104define <vscale x 4 x i16> @promote_insert_4i16(<vscale x 4 x i16> %a, i16 %elt) { 105; CHECK-LABEL: promote_insert_4i16: 106; CHECK: // %bb.0: 107; CHECK-NEXT: mov w8, #5 108; CHECK-NEXT: index z1.s, #0, #1 109; CHECK-NEXT: mov z2.s, w8 110; CHECK-NEXT: ptrue p0.s 111; CHECK-NEXT: cmpeq p0.s, p0/z, z1.s, z2.s 112; CHECK-NEXT: mov z0.s, p0/m, w0 113; CHECK-NEXT: ret 114 %ins = insertelement <vscale x 4 x i16> %a, i16 %elt, i64 5 115 ret <vscale x 4 x i16> %ins 116} 117 118; In this test, the index is small enough that we know it will be in the 119; low half of the vector and there is no need to go through the stack as 120; done in the remaining tests 121define <vscale x 32 x i8> @split_insert_32i8(<vscale x 32 x i8> %a, i8 %elt) { 122; CHECK-LABEL: split_insert_32i8: 123; CHECK: // %bb.0: 124; CHECK-NEXT: mov w8, #3 125; CHECK-NEXT: index z2.b, #0, #1 126; CHECK-NEXT: mov z3.b, w8 127; CHECK-NEXT: ptrue p0.b 128; CHECK-NEXT: cmpeq p0.b, p0/z, z2.b, z3.b 129; CHECK-NEXT: mov z0.b, p0/m, w0 130; CHECK-NEXT: ret 131 %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 3 132 ret <vscale x 32 x i8> %ins 133} 134 135define <vscale x 32 x i16> @split_insert_32i16(<vscale x 32 x i16> %a, i16 %elt) { 136; CHECK-LABEL: split_insert_32i16: 137; CHECK: // %bb.0: 138; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 139; CHECK-NEXT: addvl sp, sp, #-4 140; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG 141; CHECK-NEXT: .cfi_offset w29, -16 142; CHECK-NEXT: rdvl x10, #2 143; CHECK-NEXT: sub x10, x10, #1 // =1 144; CHECK-NEXT: mov w9, #128 145; CHECK-NEXT: cmp x10, #128 // =128 146; CHECK-NEXT: ptrue p0.h 147; CHECK-NEXT: mov x8, sp 148; CHECK-NEXT: csel x9, x10, x9, lo 149; CHECK-NEXT: st1h { z3.h }, p0, [x8, #3, mul vl] 150; CHECK-NEXT: st1h { z2.h }, p0, [x8, #2, mul vl] 151; CHECK-NEXT: st1h { z1.h }, p0, [x8, #1, mul vl] 152; CHECK-NEXT: st1h { z0.h }, p0, [sp] 153; CHECK-NEXT: strh w0, [x8, x9, lsl #1] 154; CHECK-NEXT: ld1h { z1.h }, p0/z, [x8, #1, mul vl] 155; CHECK-NEXT: ld1h { z2.h }, p0/z, [x8, #2, mul vl] 156; CHECK-NEXT: ld1h { z3.h }, p0/z, [x8, #3, mul vl] 157; CHECK-NEXT: ld1h { z0.h }, p0/z, [sp] 158; CHECK-NEXT: addvl sp, sp, #4 159; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 160; CHECK-NEXT: ret 161 %ins = insertelement <vscale x 32 x i16> %a, i16 %elt, i64 128 162 ret <vscale x 32 x i16> %ins 163} 164 165define <vscale x 8 x i32> @split_insert_8i32(<vscale x 8 x i32> %a, i32 %elt) { 166; CHECK-LABEL: split_insert_8i32: 167; CHECK: // %bb.0: 168; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill 169; CHECK-NEXT: addvl sp, sp, #-2 170; CHECK-NEXT: .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG 171; CHECK-NEXT: .cfi_offset w29, -16 172; CHECK-NEXT: mov w9, #16960 173; CHECK-NEXT: cnth x10 174; CHECK-NEXT: movk w9, #15, lsl #16 175; CHECK-NEXT: sub x10, x10, #1 // =1 176; CHECK-NEXT: cmp x10, x9 177; CHECK-NEXT: ptrue p0.s 178; CHECK-NEXT: mov x8, sp 179; CHECK-NEXT: csel x9, x10, x9, lo 180; CHECK-NEXT: st1w { z1.s }, p0, [x8, #1, mul vl] 181; CHECK-NEXT: st1w { z0.s }, p0, [sp] 182; CHECK-NEXT: str w0, [x8, x9, lsl #2] 183; CHECK-NEXT: ld1w { z1.s }, p0/z, [x8, #1, mul vl] 184; CHECK-NEXT: ld1w { z0.s }, p0/z, [sp] 185; CHECK-NEXT: addvl sp, sp, #2 186; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload 187; CHECK-NEXT: ret 188 %ins = insertelement <vscale x 8 x i32> %a, i32 %elt, i64 1000000 189 ret <vscale x 8 x i32> %ins 190} 191