1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
|
; RUN: llc < %s -march=arm64 -aarch64-stp-suppress=false -verify-machineinstrs -mcpu=cyclone | FileCheck %s
; RUN: llc < %s -march=arm64 -aarch64-unscaled-mem-op=true\
; RUN: -verify-machineinstrs -mcpu=cyclone | FileCheck -check-prefix=STUR_CHK %s
; CHECK: stp_int
; CHECK: stp w0, w1, [x2]
define void @stp_int(i32 %a, i32 %b, i32* nocapture %p) nounwind {
store i32 %a, i32* %p, align 4
%add.ptr = getelementptr inbounds i32, i32* %p, i64 1
store i32 %b, i32* %add.ptr, align 4
ret void
}
; CHECK: stp_long
; CHECK: stp x0, x1, [x2]
define void @stp_long(i64 %a, i64 %b, i64* nocapture %p) nounwind {
store i64 %a, i64* %p, align 8
%add.ptr = getelementptr inbounds i64, i64* %p, i64 1
store i64 %b, i64* %add.ptr, align 8
ret void
}
; CHECK: stp_float
; CHECK: stp s0, s1, [x0]
define void @stp_float(float %a, float %b, float* nocapture %p) nounwind {
store float %a, float* %p, align 4
%add.ptr = getelementptr inbounds float, float* %p, i64 1
store float %b, float* %add.ptr, align 4
ret void
}
; CHECK: stp_double
; CHECK: stp d0, d1, [x0]
define void @stp_double(double %a, double %b, double* nocapture %p) nounwind {
store double %a, double* %p, align 8
%add.ptr = getelementptr inbounds double, double* %p, i64 1
store double %b, double* %add.ptr, align 8
ret void
}
; Test the load/store optimizer---combine ldurs into a ldp, if appropriate
define void @stur_int(i32 %a, i32 %b, i32* nocapture %p) nounwind {
; STUR_CHK: stur_int
; STUR_CHK: stp w{{[0-9]+}}, {{w[0-9]+}}, [x{{[0-9]+}}, #-8]
; STUR_CHK-NEXT: ret
%p1 = getelementptr inbounds i32, i32* %p, i32 -1
store i32 %a, i32* %p1, align 2
%p2 = getelementptr inbounds i32, i32* %p, i32 -2
store i32 %b, i32* %p2, align 2
ret void
}
define void @stur_long(i64 %a, i64 %b, i64* nocapture %p) nounwind {
; STUR_CHK: stur_long
; STUR_CHK: stp x{{[0-9]+}}, {{x[0-9]+}}, [x{{[0-9]+}}, #-16]
; STUR_CHK-NEXT: ret
%p1 = getelementptr inbounds i64, i64* %p, i32 -1
store i64 %a, i64* %p1, align 2
%p2 = getelementptr inbounds i64, i64* %p, i32 -2
store i64 %b, i64* %p2, align 2
ret void
}
define void @stur_float(float %a, float %b, float* nocapture %p) nounwind {
; STUR_CHK: stur_float
; STUR_CHK: stp s{{[0-9]+}}, {{s[0-9]+}}, [x{{[0-9]+}}, #-8]
; STUR_CHK-NEXT: ret
%p1 = getelementptr inbounds float, float* %p, i32 -1
store float %a, float* %p1, align 2
%p2 = getelementptr inbounds float, float* %p, i32 -2
store float %b, float* %p2, align 2
ret void
}
define void @stur_double(double %a, double %b, double* nocapture %p) nounwind {
; STUR_CHK: stur_double
; STUR_CHK: stp d{{[0-9]+}}, {{d[0-9]+}}, [x{{[0-9]+}}, #-16]
; STUR_CHK-NEXT: ret
%p1 = getelementptr inbounds double, double* %p, i32 -1
store double %a, double* %p1, align 2
%p2 = getelementptr inbounds double, double* %p, i32 -2
store double %b, double* %p2, align 2
ret void
}
define void @splat_v4i32(i32 %v, i32 *%p) {
entry:
; CHECK-LABEL: splat_v4i32
; CHECK-DAG: stp w0, w0, [x1]
; CHECK-DAG: stp w0, w0, [x1, #8]
; CHECK: ret
%p17 = insertelement <4 x i32> undef, i32 %v, i32 0
%p18 = insertelement <4 x i32> %p17, i32 %v, i32 1
%p19 = insertelement <4 x i32> %p18, i32 %v, i32 2
%p20 = insertelement <4 x i32> %p19, i32 %v, i32 3
%p21 = bitcast i32* %p to <4 x i32>*
store <4 x i32> %p20, <4 x i32>* %p21, align 4
ret void
}
; Read of %b to compute %tmp2 shouldn't prevent formation of stp
; CHECK-LABEL: stp_int_rar_hazard
; CHECK: stp w0, w1, [x2]
; CHECK: ldr [[REG:w[0-9]+]], [x2, #8]
; CHECK: add w0, [[REG]], w1
; CHECK: ret
define i32 @stp_int_rar_hazard(i32 %a, i32 %b, i32* nocapture %p) nounwind {
store i32 %a, i32* %p, align 4
%ld.ptr = getelementptr inbounds i32, i32* %p, i64 2
%tmp = load i32, i32* %ld.ptr, align 4
%tmp2 = add i32 %tmp, %b
%add.ptr = getelementptr inbounds i32, i32* %p, i64 1
store i32 %b, i32* %add.ptr, align 4
ret i32 %tmp2
}
; Read of %b to compute %tmp2 shouldn't prevent formation of stp
; CHECK-LABEL: stp_int_rar_hazard_after
; CHECK: ldr [[REG:w[0-9]+]], [x3, #4]
; CHECK: add w0, [[REG]], w2
; CHECK: stp w1, w2, [x3]
; CHECK: ret
define i32 @stp_int_rar_hazard_after(i32 %w0, i32 %a, i32 %b, i32* nocapture %p) nounwind {
store i32 %a, i32* %p, align 4
%ld.ptr = getelementptr inbounds i32, i32* %p, i64 1
%tmp = load i32, i32* %ld.ptr, align 4
%tmp2 = add i32 %tmp, %b
%add.ptr = getelementptr inbounds i32, i32* %p, i64 1
store i32 %b, i32* %add.ptr, align 4
ret i32 %tmp2
}
|
