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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
|
; RUN: llc -verify-machineinstrs -o - %s -mtriple=aarch64-linux-gnu | FileCheck %s
; RUN: llc -verify-machineinstrs < %s -mtriple=aarch64-none-linux-gnu -mattr=-fp-armv8 | FileCheck --check-prefix=CHECK-NOFP %s
@var_8bit = dso_local global i8 0
@var_16bit = dso_local global i16 0
@var_32bit = dso_local global i32 0
@var_64bit = dso_local global i64 0
@var_float = dso_local global float 0.0
@var_double = dso_local global double 0.0
@varptr = dso_local global i8* null
define dso_local void @ldst_8bit() {
; CHECK-LABEL: ldst_8bit:
; No architectural support for loads to 16-bit or 8-bit since we
; promote i8 during lowering.
%addr_8bit = load i8*, i8** @varptr
; match a sign-extending load 8-bit -> 32-bit
%addr_sext32 = getelementptr i8, i8* %addr_8bit, i64 -256
%val8_sext32 = load volatile i8, i8* %addr_sext32
%val32_signed = sext i8 %val8_sext32 to i32
store volatile i32 %val32_signed, i32* @var_32bit
; CHECK: ldursb {{w[0-9]+}}, [{{x[0-9]+}}, #-256]
; match a zero-extending load volatile 8-bit -> 32-bit
%addr_zext32 = getelementptr i8, i8* %addr_8bit, i64 -12
%val8_zext32 = load volatile i8, i8* %addr_zext32
%val32_unsigned = zext i8 %val8_zext32 to i32
store volatile i32 %val32_unsigned, i32* @var_32bit
; CHECK: ldurb {{w[0-9]+}}, [{{x[0-9]+}}, #-12]
; match an any-extending load volatile 8-bit -> 32-bit
%addr_anyext = getelementptr i8, i8* %addr_8bit, i64 -1
%val8_anyext = load volatile i8, i8* %addr_anyext
%newval8 = add i8 %val8_anyext, 1
store volatile i8 %newval8, i8* @var_8bit
; CHECK: ldurb {{w[0-9]+}}, [{{x[0-9]+}}, #-1]
; match a sign-extending load volatile 8-bit -> 64-bit
%addr_sext64 = getelementptr i8, i8* %addr_8bit, i64 -5
%val8_sext64 = load volatile i8, i8* %addr_sext64
%val64_signed = sext i8 %val8_sext64 to i64
store volatile i64 %val64_signed, i64* @var_64bit
; CHECK: ldursb {{x[0-9]+}}, [{{x[0-9]+}}, #-5]
; match a zero-extending load volatile 8-bit -> 64-bit.
; This uses the fact that ldrb w0, [x0] will zero out the high 32-bits
; of x0 so it's identical to load volatileing to 32-bits.
%addr_zext64 = getelementptr i8, i8* %addr_8bit, i64 -9
%val8_zext64 = load volatile i8, i8* %addr_zext64
%val64_unsigned = zext i8 %val8_zext64 to i64
store volatile i64 %val64_unsigned, i64* @var_64bit
; CHECK: ldurb {{w[0-9]+}}, [{{x[0-9]+}}, #-9]
; truncating store volatile 32-bits to 8-bits
%addr_trunc32 = getelementptr i8, i8* %addr_8bit, i64 -256
%val32 = load volatile i32, i32* @var_32bit
%val8_trunc32 = trunc i32 %val32 to i8
store volatile i8 %val8_trunc32, i8* %addr_trunc32
; CHECK: sturb {{w[0-9]+}}, [{{x[0-9]+}}, #-256]
; truncating store volatile 64-bits to 8-bits
%addr_trunc64 = getelementptr i8, i8* %addr_8bit, i64 -1
%val64 = load volatile i64, i64* @var_64bit
%val8_trunc64 = trunc i64 %val64 to i8
store volatile i8 %val8_trunc64, i8* %addr_trunc64
; CHECK: sturb {{w[0-9]+}}, [{{x[0-9]+}}, #-1]
ret void
}
define dso_local void @ldst_16bit() {
; CHECK-LABEL: ldst_16bit:
; No architectural support for loads to 16-bit or 16-bit since we
; promote i16 during lowering.
%addr_8bit = load i8*, i8** @varptr
; match a sign-extending load 16-bit -> 32-bit
%addr8_sext32 = getelementptr i8, i8* %addr_8bit, i64 -256
%addr_sext32 = bitcast i8* %addr8_sext32 to i16*
%val16_sext32 = load volatile i16, i16* %addr_sext32
%val32_signed = sext i16 %val16_sext32 to i32
store volatile i32 %val32_signed, i32* @var_32bit
; CHECK: ldursh {{w[0-9]+}}, [{{x[0-9]+}}, #-256]
; match a zero-extending load volatile 16-bit -> 32-bit. With offset that would be unaligned.
%addr8_zext32 = getelementptr i8, i8* %addr_8bit, i64 15
%addr_zext32 = bitcast i8* %addr8_zext32 to i16*
%val16_zext32 = load volatile i16, i16* %addr_zext32
%val32_unsigned = zext i16 %val16_zext32 to i32
store volatile i32 %val32_unsigned, i32* @var_32bit
; CHECK: ldurh {{w[0-9]+}}, [{{x[0-9]+}}, #15]
; match an any-extending load volatile 16-bit -> 32-bit
%addr8_anyext = getelementptr i8, i8* %addr_8bit, i64 -1
%addr_anyext = bitcast i8* %addr8_anyext to i16*
%val16_anyext = load volatile i16, i16* %addr_anyext
%newval16 = add i16 %val16_anyext, 1
store volatile i16 %newval16, i16* @var_16bit
; CHECK: ldurh {{w[0-9]+}}, [{{x[0-9]+}}, #-1]
; match a sign-extending load volatile 16-bit -> 64-bit
%addr8_sext64 = getelementptr i8, i8* %addr_8bit, i64 -5
%addr_sext64 = bitcast i8* %addr8_sext64 to i16*
%val16_sext64 = load volatile i16, i16* %addr_sext64
%val64_signed = sext i16 %val16_sext64 to i64
store volatile i64 %val64_signed, i64* @var_64bit
; CHECK: ldursh {{x[0-9]+}}, [{{x[0-9]+}}, #-5]
; match a zero-extending load volatile 16-bit -> 64-bit.
; This uses the fact that ldrb w0, [x0] will zero out the high 32-bits
; of x0 so it's identical to load volatileing to 32-bits.
%addr8_zext64 = getelementptr i8, i8* %addr_8bit, i64 9
%addr_zext64 = bitcast i8* %addr8_zext64 to i16*
%val16_zext64 = load volatile i16, i16* %addr_zext64
%val64_unsigned = zext i16 %val16_zext64 to i64
store volatile i64 %val64_unsigned, i64* @var_64bit
; CHECK: ldurh {{w[0-9]+}}, [{{x[0-9]+}}, #9]
; truncating store volatile 32-bits to 16-bits
%addr8_trunc32 = getelementptr i8, i8* %addr_8bit, i64 -256
%addr_trunc32 = bitcast i8* %addr8_trunc32 to i16*
%val32 = load volatile i32, i32* @var_32bit
%val16_trunc32 = trunc i32 %val32 to i16
store volatile i16 %val16_trunc32, i16* %addr_trunc32
; CHECK: sturh {{w[0-9]+}}, [{{x[0-9]+}}, #-256]
; truncating store volatile 64-bits to 16-bits
%addr8_trunc64 = getelementptr i8, i8* %addr_8bit, i64 -1
%addr_trunc64 = bitcast i8* %addr8_trunc64 to i16*
%val64 = load volatile i64, i64* @var_64bit
%val16_trunc64 = trunc i64 %val64 to i16
store volatile i16 %val16_trunc64, i16* %addr_trunc64
; CHECK: sturh {{w[0-9]+}}, [{{x[0-9]+}}, #-1]
ret void
}
define dso_local void @ldst_32bit() {
; CHECK-LABEL: ldst_32bit:
%addr_8bit = load i8*, i8** @varptr
; Straight 32-bit load/store
%addr32_8_noext = getelementptr i8, i8* %addr_8bit, i64 1
%addr32_noext = bitcast i8* %addr32_8_noext to i32*
%val32_noext = load volatile i32, i32* %addr32_noext
store volatile i32 %val32_noext, i32* %addr32_noext
; CHECK: ldur {{w[0-9]+}}, [{{x[0-9]+}}, #1]
; CHECK: stur {{w[0-9]+}}, [{{x[0-9]+}}, #1]
; Zero-extension to 64-bits
%addr32_8_zext = getelementptr i8, i8* %addr_8bit, i64 -256
%addr32_zext = bitcast i8* %addr32_8_zext to i32*
%val32_zext = load volatile i32, i32* %addr32_zext
%val64_unsigned = zext i32 %val32_zext to i64
store volatile i64 %val64_unsigned, i64* @var_64bit
; CHECK: ldur {{w[0-9]+}}, [{{x[0-9]+}}, #-256]
; CHECK: str {{x[0-9]+}}, [{{x[0-9]+}}, {{#?}}:lo12:var_64bit]
; Sign-extension to 64-bits
%addr32_8_sext = getelementptr i8, i8* %addr_8bit, i64 -12
%addr32_sext = bitcast i8* %addr32_8_sext to i32*
%val32_sext = load volatile i32, i32* %addr32_sext
%val64_signed = sext i32 %val32_sext to i64
store volatile i64 %val64_signed, i64* @var_64bit
; CHECK: ldursw {{x[0-9]+}}, [{{x[0-9]+}}, #-12]
; CHECK: str {{x[0-9]+}}, [{{x[0-9]+}}, {{#?}}:lo12:var_64bit]
; Truncation from 64-bits
%addr64_8_trunc = getelementptr i8, i8* %addr_8bit, i64 255
%addr64_trunc = bitcast i8* %addr64_8_trunc to i64*
%addr32_8_trunc = getelementptr i8, i8* %addr_8bit, i64 -20
%addr32_trunc = bitcast i8* %addr32_8_trunc to i32*
%val64_trunc = load volatile i64, i64* %addr64_trunc
%val32_trunc = trunc i64 %val64_trunc to i32
store volatile i32 %val32_trunc, i32* %addr32_trunc
; CHECK: ldur {{x[0-9]+}}, [{{x[0-9]+}}, #255]
; CHECK: stur {{w[0-9]+}}, [{{x[0-9]+}}, #-20]
ret void
}
define dso_local void @ldst_float() {
; CHECK-LABEL: ldst_float:
%addr_8bit = load i8*, i8** @varptr
%addrfp_8 = getelementptr i8, i8* %addr_8bit, i64 -5
%addrfp = bitcast i8* %addrfp_8 to float*
%valfp = load volatile float, float* %addrfp
; CHECK: ldur {{s[0-9]+}}, [{{x[0-9]+}}, #-5]
; CHECK-NOFP-NOT: ldur {{s[0-9]+}},
store volatile float %valfp, float* %addrfp
; CHECK: stur {{s[0-9]+}}, [{{x[0-9]+}}, #-5]
; CHECK-NOFP-NOT: stur {{s[0-9]+}},
ret void
}
define dso_local void @ldst_double() {
; CHECK-LABEL: ldst_double:
%addr_8bit = load i8*, i8** @varptr
%addrfp_8 = getelementptr i8, i8* %addr_8bit, i64 4
%addrfp = bitcast i8* %addrfp_8 to double*
%valfp = load volatile double, double* %addrfp
; CHECK: ldur {{d[0-9]+}}, [{{x[0-9]+}}, #4]
; CHECK-NOFP-NOT: ldur {{d[0-9]+}},
store volatile double %valfp, double* %addrfp
; CHECK: stur {{d[0-9]+}}, [{{x[0-9]+}}, #4]
; CHECK-NOFP-NOT: stur {{d[0-9]+}},
ret void
}
|
