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
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
| //===----------------------Hexagon builtin routine ------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// Functions that implement common sequences in function prologues and epilogues
// used to save code size
.macro FUNCTION_BEGIN name
.text
.globl \name
.type \name, @function
.falign
\name:
.endm
.macro FUNCTION_END name
.size \name, . - \name
.endm
.macro FALLTHROUGH_TAIL_CALL name0 name1
.size \name0, . - \name0
.globl \name1
.type \name1, @function
.falign
\name1:
.endm
// Save r27:26 at fp+#-8, r25:24 at fp+#-16, r23:22 at fp+#-24, r21:20 at
// fp+#-32, r19:18 at fp+#-40, and r17:16 at fp+#-48.
// The compiler knows that the __save_* functions clobber LR. No other
// registers should be used without informing the compiler.
// Since we can only issue one store per packet, we don't hurt performance by
// simply jumping to the right point in this sequence of stores.
FUNCTION_BEGIN __save_r27_through_r16
memd(fp+#-48) = r17:16
FALLTHROUGH_TAIL_CALL __save_r27_through_r16 __save_r27_through_r18
memd(fp+#-40) = r19:18
FALLTHROUGH_TAIL_CALL __save_r27_through_r18 __save_r27_through_r20
memd(fp+#-32) = r21:20
FALLTHROUGH_TAIL_CALL __save_r27_through_r20 __save_r27_through_r22
memd(fp+#-24) = r23:22
FALLTHROUGH_TAIL_CALL __save_r27_through_r22 __save_r27_through_r24
memd(fp+#-16) = r25:24
{
memd(fp+#-8) = r27:26
jumpr lr
}
FUNCTION_END __save_r27_through_r24
// For each of the *_before_sibcall functions, jumpr lr is executed in parallel
// with deallocframe. That way, the return gets the old value of lr, which is
// where these functions need to return, and at the same time, lr gets the value
// it needs going into the sibcall.
FUNCTION_BEGIN __restore_r27_through_r20_and_deallocframe_before_sibcall
{
r21:20 = memd(fp+#-32)
r23:22 = memd(fp+#-24)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r20_and_deallocframe_before_sibcall __restore_r27_through_r24_and_deallocframe_before_sibcall
{
r25:24 = memd(fp+#-16)
jump __restore_r27_through_r26_and_deallocframe_before_sibcall
}
FUNCTION_END __restore_r27_through_r24_and_deallocframe_before_sibcall
FUNCTION_BEGIN __restore_r27_through_r16_and_deallocframe_before_sibcall
r17:16 = memd(fp+#-48)
FALLTHROUGH_TAIL_CALL __restore_r27_through_r16_and_deallocframe_before_sibcall __restore_r27_through_r18_and_deallocframe_before_sibcall
{
r19:18 = memd(fp+#-40)
r21:20 = memd(fp+#-32)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r18_and_deallocframe_before_sibcall __restore_r27_through_r22_and_deallocframe_before_sibcall
{
r23:22 = memd(fp+#-24)
r25:24 = memd(fp+#-16)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r22_and_deallocframe_before_sibcall __restore_r27_through_r26_and_deallocframe_before_sibcall
{
r27:26 = memd(fp+#-8)
deallocframe
jumpr lr
}
FUNCTION_END __restore_r27_through_r26_and_deallocframe_before_sibcall
// Here we use the extra load bandwidth to restore LR early, allowing the return
// to occur in parallel with the deallocframe.
FUNCTION_BEGIN __restore_r27_through_r16_and_deallocframe
{
r17:16 = memd(fp+#-48)
r19:18 = memd(fp+#-40)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r16_and_deallocframe __restore_r27_through_r20_and_deallocframe
{
r21:20 = memd(fp+#-32)
r23:22 = memd(fp+#-24)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r20_and_deallocframe __restore_r27_through_r24_and_deallocframe
{
lr = memw(fp+#4)
r25:24 = memd(fp+#-16)
}
{
r27:26 = memd(fp+#-8)
deallocframe
jumpr lr
}
FUNCTION_END __restore_r27_through_r24_and_deallocframe
// Here the load bandwidth is maximized for all three functions.
FUNCTION_BEGIN __restore_r27_through_r18_and_deallocframe
{
r19:18 = memd(fp+#-40)
r21:20 = memd(fp+#-32)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r18_and_deallocframe __restore_r27_through_r22_and_deallocframe
{
r23:22 = memd(fp+#-24)
r25:24 = memd(fp+#-16)
}
FALLTHROUGH_TAIL_CALL __restore_r27_through_r22_and_deallocframe __restore_r27_through_r26_and_deallocframe
{
r27:26 = memd(fp+#-8)
deallocframe
}
jumpr lr
FUNCTION_END __restore_r27_through_r26_and_deallocframe
|