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:    .cfi_def_cfa_offset 16
24; CHECK-NEXT:    .cfi_offset w29, -16
25; CHECK-NEXT:    addvl sp, sp, #-2
26; CHECK-NEXT:    .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG
27; CHECK-NEXT:    mov x8, #-1
28; CHECK-NEXT:    mov x9, sp
29; CHECK-NEXT:    ptrue p0.b
30; CHECK-NEXT:    st1b { z1.b }, p0, [sp, #1, mul vl]
31; CHECK-NEXT:    st1b { z0.b }, p0, [sp]
32; CHECK-NEXT:    addvl x8, x8, #2
33; CHECK-NEXT:    cmp x1, x8
34; CHECK-NEXT:    csel x8, x1, x8, lo
35; CHECK-NEXT:    strb w0, [x9, x8]
36; CHECK-NEXT:    ld1b { z0.b }, p0/z, [sp]
37; CHECK-NEXT:    ld1b { z1.b }, p0/z, [sp, #1, mul vl]
38; CHECK-NEXT:    addvl sp, sp, #2
39; CHECK-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
40; CHECK-NEXT:    ret
41  %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 %idx
42  ret <vscale x 32 x i8> %ins
43}
44
45define <vscale x 8 x float> @split_insert_8f32_idx(<vscale x 8 x float> %a, float %elt, i64 %idx) {
46; CHECK-LABEL: split_insert_8f32_idx:
47; CHECK:       // %bb.0:
48; CHECK-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
49; CHECK-NEXT:    .cfi_def_cfa_offset 16
50; CHECK-NEXT:    .cfi_offset w29, -16
51; CHECK-NEXT:    addvl sp, sp, #-2
52; CHECK-NEXT:    .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG
53; CHECK-NEXT:    cnth x8
54; CHECK-NEXT:    mov x9, sp
55; CHECK-NEXT:    sub x8, x8, #1
56; CHECK-NEXT:    ptrue p0.s
57; CHECK-NEXT:    cmp x0, x8
58; CHECK-NEXT:    st1w { z1.s }, p0, [sp, #1, mul vl]
59; CHECK-NEXT:    csel x8, x0, x8, lo
60; CHECK-NEXT:    st1w { z0.s }, p0, [sp]
61; CHECK-NEXT:    str s2, [x9, x8, lsl #2]
62; CHECK-NEXT:    ld1w { z0.s }, p0/z, [sp]
63; CHECK-NEXT:    ld1w { z1.s }, p0/z, [sp, #1, mul vl]
64; CHECK-NEXT:    addvl sp, sp, #2
65; CHECK-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
66; CHECK-NEXT:    ret
67  %ins = insertelement <vscale x 8 x float> %a, float %elt, i64 %idx
68  ret <vscale x 8 x float> %ins
69}
70
71define <vscale x 8 x i64> @split_insert_8i64_idx(<vscale x 8 x i64> %a, i64 %elt, i64 %idx) {
72; CHECK-LABEL: split_insert_8i64_idx:
73; CHECK:       // %bb.0:
74; CHECK-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
75; CHECK-NEXT:    .cfi_def_cfa_offset 16
76; CHECK-NEXT:    .cfi_offset w29, -16
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:    cnth x8
80; CHECK-NEXT:    mov x9, sp
81; CHECK-NEXT:    sub x8, x8, #1
82; CHECK-NEXT:    ptrue p0.d
83; CHECK-NEXT:    cmp x1, x8
84; CHECK-NEXT:    st1d { z3.d }, p0, [sp, #3, mul vl]
85; CHECK-NEXT:    csel x8, x1, x8, lo
86; CHECK-NEXT:    st1d { z2.d }, p0, [sp, #2, mul vl]
87; CHECK-NEXT:    st1d { z1.d }, p0, [sp, #1, mul vl]
88; CHECK-NEXT:    st1d { z0.d }, p0, [sp]
89; CHECK-NEXT:    str x0, [x9, x8, lsl #3]
90; CHECK-NEXT:    ld1d { z0.d }, p0/z, [sp]
91; CHECK-NEXT:    ld1d { z1.d }, p0/z, [sp, #1, mul vl]
92; CHECK-NEXT:    ld1d { z2.d }, p0/z, [sp, #2, mul vl]
93; CHECK-NEXT:    ld1d { z3.d }, p0/z, [sp, #3, mul vl]
94; CHECK-NEXT:    addvl sp, sp, #4
95; CHECK-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
96; CHECK-NEXT:    ret
97  %ins = insertelement <vscale x 8 x i64> %a, i64 %elt, i64 %idx
98  ret <vscale x 8 x i64> %ins
99}
100
101; INSERT VECTOR ELT, CONSTANT IDX
102
103define <vscale x 4 x i16> @promote_insert_4i16(<vscale x 4 x i16> %a, i16 %elt) {
104; CHECK-LABEL: promote_insert_4i16:
105; CHECK:       // %bb.0:
106; CHECK-NEXT:    mov w8, #5
107; CHECK-NEXT:    index z2.s, #0, #1
108; CHECK-NEXT:    ptrue p0.s
109; CHECK-NEXT:    mov z1.s, w8
110; CHECK-NEXT:    cmpeq p0.s, p0/z, z2.s, z1.s
111; CHECK-NEXT:    mov z0.s, p0/m, w0
112; CHECK-NEXT:    ret
113  %ins = insertelement <vscale x 4 x i16> %a, i16 %elt, i64 5
114  ret <vscale x 4 x i16> %ins
115}
116
117; In this test, the index is small enough that we know it will be in the
118; low half of the vector and there is no need to go through the stack as
119; done in the remaining tests
120define <vscale x 32 x i8> @split_insert_32i8(<vscale x 32 x i8> %a, i8 %elt) {
121; CHECK-LABEL: split_insert_32i8:
122; CHECK:       // %bb.0:
123; CHECK-NEXT:    mov w8, #3
124; CHECK-NEXT:    index z3.b, #0, #1
125; CHECK-NEXT:    ptrue p0.b
126; CHECK-NEXT:    mov z2.b, w8
127; CHECK-NEXT:    cmpeq p0.b, p0/z, z3.b, z2.b
128; CHECK-NEXT:    mov z0.b, p0/m, w0
129; CHECK-NEXT:    ret
130  %ins = insertelement <vscale x 32 x i8> %a, i8 %elt, i64 3
131  ret <vscale x 32 x i8> %ins
132}
133
134define <vscale x 32 x i16> @split_insert_32i16(<vscale x 32 x i16> %a, i16 %elt) {
135; CHECK-LABEL: split_insert_32i16:
136; CHECK:       // %bb.0:
137; CHECK-NEXT:    str x29, [sp, #-16]! // 8-byte Folded Spill
138; CHECK-NEXT:    .cfi_def_cfa_offset 16
139; CHECK-NEXT:    .cfi_offset w29, -16
140; CHECK-NEXT:    addvl sp, sp, #-4
141; CHECK-NEXT:    .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x20, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 32 * VG
142; CHECK-NEXT:    mov x8, #-1
143; CHECK-NEXT:    mov w9, #128
144; CHECK-NEXT:    ptrue p0.h
145; CHECK-NEXT:    st1h { z3.h }, p0, [sp, #3, mul vl]
146; CHECK-NEXT:    st1h { z2.h }, p0, [sp, #2, mul vl]
147; CHECK-NEXT:    addvl x8, x8, #2
148; CHECK-NEXT:    st1h { z1.h }, p0, [sp, #1, mul vl]
149; CHECK-NEXT:    cmp x8, #128
150; CHECK-NEXT:    st1h { z0.h }, p0, [sp]
151; CHECK-NEXT:    csel x8, x8, x9, lo
152; CHECK-NEXT:    mov x9, sp
153; CHECK-NEXT:    strh w0, [x9, x8, lsl #1]
154; CHECK-NEXT:    ld1h { z0.h }, p0/z, [sp]
155; CHECK-NEXT:    ld1h { z1.h }, p0/z, [sp, #1, mul vl]
156; CHECK-NEXT:    ld1h { z2.h }, p0/z, [sp, #2, mul vl]
157; CHECK-NEXT:    ld1h { z3.h }, p0/z, [sp, #3, mul vl]
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:    .cfi_def_cfa_offset 16
170; CHECK-NEXT:    .cfi_offset w29, -16
171; CHECK-NEXT:    addvl sp, sp, #-2
172; CHECK-NEXT:    .cfi_escape 0x0f, 0x0c, 0x8f, 0x00, 0x11, 0x10, 0x22, 0x11, 0x10, 0x92, 0x2e, 0x00, 0x1e, 0x22 // sp + 16 + 16 * VG
173; CHECK-NEXT:    cnth x8
174; CHECK-NEXT:    mov w9, #16960
175; CHECK-NEXT:    movk w9, #15, lsl #16
176; CHECK-NEXT:    sub x8, x8, #1
177; CHECK-NEXT:    cmp x8, x9
178; CHECK-NEXT:    ptrue p0.s
179; CHECK-NEXT:    csel x8, x8, x9, lo
180; CHECK-NEXT:    mov x9, sp
181; CHECK-NEXT:    st1w { z1.s }, p0, [sp, #1, mul vl]
182; CHECK-NEXT:    st1w { z0.s }, p0, [sp]
183; CHECK-NEXT:    str w0, [x9, x8, lsl #2]
184; CHECK-NEXT:    ld1w { z0.s }, p0/z, [sp]
185; CHECK-NEXT:    ld1w { z1.s }, p0/z, [sp, #1, mul vl]
186; CHECK-NEXT:    addvl sp, sp, #2
187; CHECK-NEXT:    ldr x29, [sp], #16 // 8-byte Folded Reload
188; CHECK-NEXT:    ret
189  %ins = insertelement <vscale x 8 x i32> %a, i32 %elt, i64 1000000
190  ret <vscale x 8 x i32> %ins
191}
192