1; RUN: llc < %s -mtriple=arm64-eabi -aarch64-enable-stp-suppress=false -verify-machineinstrs -mcpu=cyclone | FileCheck %s
2
3; CHECK-LABEL: stp_int
4; CHECK: stp w0, w1, [x2]
5define void @stp_int(i32 %a, i32 %b, i32* nocapture %p) nounwind {
6  store i32 %a, i32* %p, align 4
7  %add.ptr = getelementptr inbounds i32, i32* %p, i64 1
8  store i32 %b, i32* %add.ptr, align 4
9  ret void
10}
11
12; CHECK-LABEL: stp_long
13; CHECK: stp x0, x1, [x2]
14define void @stp_long(i64 %a, i64 %b, i64* nocapture %p) nounwind {
15  store i64 %a, i64* %p, align 8
16  %add.ptr = getelementptr inbounds i64, i64* %p, i64 1
17  store i64 %b, i64* %add.ptr, align 8
18  ret void
19}
20
21; CHECK-LABEL: stp_float
22; CHECK: stp s0, s1, [x0]
23define void @stp_float(float %a, float %b, float* nocapture %p) nounwind {
24  store float %a, float* %p, align 4
25  %add.ptr = getelementptr inbounds float, float* %p, i64 1
26  store float %b, float* %add.ptr, align 4
27  ret void
28}
29
30; CHECK-LABEL: stp_double
31; CHECK: stp d0, d1, [x0]
32define void @stp_double(double %a, double %b, double* nocapture %p) nounwind {
33  store double %a, double* %p, align 8
34  %add.ptr = getelementptr inbounds double, double* %p, i64 1
35  store double %b, double* %add.ptr, align 8
36  ret void
37}
38
39; CHECK-LABEL: stp_doublex2
40; CHECK: stp q0, q1, [x0]
41define void @stp_doublex2(<2 x double> %a, <2 x double> %b, <2 x double>* nocapture %p) nounwind {
42  store <2 x double> %a, <2 x double>* %p, align 16
43  %add.ptr = getelementptr inbounds <2 x double>, <2 x double>* %p, i64 1
44  store <2 x double> %b, <2 x double>* %add.ptr, align 16
45  ret void
46}
47
48; Test the load/store optimizer---combine ldurs into a ldp, if appropriate
49define void @stur_int(i32 %a, i32 %b, i32* nocapture %p) nounwind {
50; CHECK-LABEL: stur_int
51; CHECK: stp w{{[0-9]+}}, {{w[0-9]+}}, [x{{[0-9]+}}, #-8]
52; CHECK-NEXT: ret
53  %p1 = getelementptr inbounds i32, i32* %p, i32 -1
54  store i32 %a, i32* %p1, align 2
55  %p2 = getelementptr inbounds i32, i32* %p, i32 -2
56  store i32 %b, i32* %p2, align 2
57  ret void
58}
59
60define void @stur_long(i64 %a, i64 %b, i64* nocapture %p) nounwind {
61; CHECK-LABEL: stur_long
62; CHECK: stp x{{[0-9]+}}, {{x[0-9]+}}, [x{{[0-9]+}}, #-16]
63; CHECK-NEXT: ret
64  %p1 = getelementptr inbounds i64, i64* %p, i32 -1
65  store i64 %a, i64* %p1, align 2
66  %p2 = getelementptr inbounds i64, i64* %p, i32 -2
67  store i64 %b, i64* %p2, align 2
68  ret void
69}
70
71define void @stur_float(float %a, float %b, float* nocapture %p) nounwind {
72; CHECK-LABEL: stur_float
73; CHECK: stp s{{[0-9]+}}, {{s[0-9]+}}, [x{{[0-9]+}}, #-8]
74; CHECK-NEXT: ret
75  %p1 = getelementptr inbounds float, float* %p, i32 -1
76  store float %a, float* %p1, align 2
77  %p2 = getelementptr inbounds float, float* %p, i32 -2
78  store float %b, float* %p2, align 2
79  ret void
80}
81
82define void @stur_double(double %a, double %b, double* nocapture %p) nounwind {
83; CHECK-LABEL: stur_double
84; CHECK: stp d{{[0-9]+}}, {{d[0-9]+}}, [x{{[0-9]+}}, #-16]
85; CHECK-NEXT: ret
86  %p1 = getelementptr inbounds double, double* %p, i32 -1
87  store double %a, double* %p1, align 2
88  %p2 = getelementptr inbounds double, double* %p, i32 -2
89  store double %b, double* %p2, align 2
90  ret void
91}
92
93define void @stur_doublex2(<2 x double> %a, <2 x double> %b, <2 x double>* nocapture %p) nounwind {
94; CHECK-LABEL: stur_doublex2
95; CHECK: stp q{{[0-9]+}}, q{{[0-9]+}}, [x{{[0-9]+}}, #-32]
96; CHECK-NEXT: ret
97  %p1 = getelementptr inbounds <2 x double>, <2 x double>* %p, i32 -1
98  store <2 x double> %a, <2 x double>* %p1, align 2
99  %p2 = getelementptr inbounds <2 x double>, <2 x double>* %p, i32 -2
100  store <2 x double> %b, <2 x double>* %p2, align 2
101  ret void
102}
103
104define void @splat_v4i32(i32 %v, i32 *%p) {
105entry:
106
107; CHECK-LABEL: splat_v4i32
108; CHECK-DAG: dup v0.4s, w0
109; CHECK-DAG: str q0, [x1]
110; CHECK: ret
111
112  %p17 = insertelement <4 x i32> undef, i32 %v, i32 0
113  %p18 = insertelement <4 x i32> %p17, i32 %v, i32 1
114  %p19 = insertelement <4 x i32> %p18, i32 %v, i32 2
115  %p20 = insertelement <4 x i32> %p19, i32 %v, i32 3
116  %p21 = bitcast i32* %p to <4 x i32>*
117  store <4 x i32> %p20, <4 x i32>* %p21, align 4
118  ret void
119}
120
121; Check that a non-splat store that is storing a vector created by 4
122; insertelements that is not a splat vector does not get split.
123define void @nosplat_v4i32(i32 %v, i32 *%p) {
124entry:
125
126; CHECK-LABEL: nosplat_v4i32:
127; CHECK: str w0,
128; CHECK: ldr q[[REG1:[0-9]+]],
129; CHECK-DAG: mov v[[REG1]].s[1], w0
130; CHECK-DAG: mov v[[REG1]].s[2], w0
131; CHECK-DAG: mov v[[REG1]].s[3], w0
132; CHECK: str q[[REG1]], [x1]
133; CHECK: ret
134
135  %p17 = insertelement <4 x i32> undef, i32 %v, i32 %v
136  %p18 = insertelement <4 x i32> %p17, i32 %v, i32 1
137  %p19 = insertelement <4 x i32> %p18, i32 %v, i32 2
138  %p20 = insertelement <4 x i32> %p19, i32 %v, i32 3
139  %p21 = bitcast i32* %p to <4 x i32>*
140  store <4 x i32> %p20, <4 x i32>* %p21, align 4
141  ret void
142}
143
144; Check that a non-splat store that is storing a vector created by 4
145; insertelements that is not a splat vector does not get split.
146define void @nosplat2_v4i32(i32 %v, i32 *%p, <4 x i32> %vin) {
147entry:
148
149; CHECK-LABEL: nosplat2_v4i32:
150; CHECK: mov v[[REG1]].s[1], w0
151; CHECK-DAG: mov v[[REG1]].s[2], w0
152; CHECK-DAG: mov v[[REG1]].s[3], w0
153; CHECK: str q[[REG1]], [x1]
154; CHECK: ret
155
156  %p18 = insertelement <4 x i32> %vin, i32 %v, i32 1
157  %p19 = insertelement <4 x i32> %p18, i32 %v, i32 2
158  %p20 = insertelement <4 x i32> %p19, i32 %v, i32 3
159  %p21 = bitcast i32* %p to <4 x i32>*
160  store <4 x i32> %p20, <4 x i32>* %p21, align 4
161  ret void
162}
163
164; Read of %b to compute %tmp2 shouldn't prevent formation of stp
165; CHECK-LABEL: stp_int_rar_hazard
166; CHECK: ldr [[REG:w[0-9]+]], [x2, #8]
167; CHECK: add w8, [[REG]], w1
168; CHECK: stp w0, w1, [x2]
169; CHECK: ret
170define i32 @stp_int_rar_hazard(i32 %a, i32 %b, i32* nocapture %p) nounwind {
171  store i32 %a, i32* %p, align 4
172  %ld.ptr = getelementptr inbounds i32, i32* %p, i64 2
173  %tmp = load i32, i32* %ld.ptr, align 4
174  %tmp2 = add i32 %tmp, %b
175  %add.ptr = getelementptr inbounds i32, i32* %p, i64 1
176  store i32 %b, i32* %add.ptr, align 4
177  ret i32 %tmp2
178}
179
180; Read of %b to compute %tmp2 shouldn't prevent formation of stp
181; CHECK-LABEL: stp_int_rar_hazard_after
182; CHECK: ldr [[REG:w[0-9]+]], [x3, #4]
183; CHECK: add w0, [[REG]], w2
184; CHECK: stp w1, w2, [x3]
185; CHECK: ret
186define i32 @stp_int_rar_hazard_after(i32 %w0, i32 %a, i32 %b, i32* nocapture %p) nounwind {
187  store i32 %a, i32* %p, align 4
188  %ld.ptr = getelementptr inbounds i32, i32* %p, i64 1
189  %tmp = load i32, i32* %ld.ptr, align 4
190  %tmp2 = add i32 %tmp, %b
191  %add.ptr = getelementptr inbounds i32, i32* %p, i64 1
192  store i32 %b, i32* %add.ptr, align 4
193  ret i32 %tmp2
194}
195