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
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --prefix-filecheck-ir-name _ --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=infer-address-spaces %s | FileCheck %s
; Trivial optimization of generic addressing
define float @load_global_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define float @load_global_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(1)
; CHECK-NEXT: [[_TMP1:%.*]] = load float, ptr addrspace(1) [[_TMP0]], align 4
; CHECK-NEXT: ret float [[_TMP1]]
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(1)
%tmp1 = load float, ptr addrspace(1) %tmp0
ret float %tmp1
}
define float @load_constant_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define float @load_constant_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(4)
; CHECK-NEXT: [[_TMP1:%.*]] = load float, ptr addrspace(4) [[_TMP0]], align 4
; CHECK-NEXT: ret float [[_TMP1]]
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(4)
%tmp1 = load float, ptr addrspace(4) %tmp0
ret float %tmp1
}
define float @load_group_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define float @load_group_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(3)
; CHECK-NEXT: [[_TMP1:%.*]] = load float, ptr addrspace(3) [[_TMP0]], align 4
; CHECK-NEXT: ret float [[_TMP1]]
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(3)
%tmp1 = load float, ptr addrspace(3) %tmp0
ret float %tmp1
}
define float @load_private_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define float @load_private_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(5)
; CHECK-NEXT: [[_TMP1:%.*]] = load float, ptr addrspace(5) [[_TMP0]], align 4
; CHECK-NEXT: ret float [[_TMP1]]
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(5)
%tmp1 = load float, ptr addrspace(5) %tmp0
ret float %tmp1
}
define amdgpu_kernel void @store_global_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define amdgpu_kernel void @store_global_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(1)
; CHECK-NEXT: store float 0.000000e+00, ptr addrspace(1) [[_TMP0]], align 4
; CHECK-NEXT: ret void
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(1)
store float 0.0, ptr addrspace(1) %tmp0
ret void
}
define amdgpu_kernel void @store_group_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define amdgpu_kernel void @store_group_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(1)
; CHECK-NEXT: [[TMP2:%.*]] = addrspacecast ptr addrspace(1) [[TMP1]] to ptr
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[TMP2]] to ptr addrspace(3)
; CHECK-NEXT: store float 0.000000e+00, ptr addrspace(3) [[_TMP0]], align 4
; CHECK-NEXT: ret void
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(3)
store float 0.0, ptr addrspace(3) %tmp0
ret void
}
define amdgpu_kernel void @store_private_from_flat(ptr %generic_scalar) #0 {
; CHECK-LABEL: define amdgpu_kernel void @store_private_from_flat(
; CHECK-SAME: ptr [[GENERIC_SCALAR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[GENERIC_SCALAR]] to ptr addrspace(1)
; CHECK-NEXT: [[TMP2:%.*]] = addrspacecast ptr addrspace(1) [[TMP1]] to ptr
; CHECK-NEXT: [[_TMP0:%.*]] = addrspacecast ptr [[TMP2]] to ptr addrspace(5)
; CHECK-NEXT: store float 0.000000e+00, ptr addrspace(5) [[_TMP0]], align 4
; CHECK-NEXT: ret void
;
%tmp0 = addrspacecast ptr %generic_scalar to ptr addrspace(5)
store float 0.0, ptr addrspace(5) %tmp0
ret void
}
; optimized to global load/store.
define amdgpu_kernel void @load_store_global(ptr addrspace(1) nocapture %input, ptr addrspace(1) nocapture %output) #0 {
; CHECK-LABEL: define amdgpu_kernel void @load_store_global(
; CHECK-SAME: ptr addrspace(1) captures(none) [[INPUT:%.*]], ptr addrspace(1) captures(none) [[OUTPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[VAL:%.*]] = load i32, ptr addrspace(1) [[INPUT]], align 4
; CHECK-NEXT: store i32 [[VAL]], ptr addrspace(1) [[OUTPUT]], align 4
; CHECK-NEXT: ret void
;
%tmp0 = addrspacecast ptr addrspace(1) %input to ptr
%tmp1 = addrspacecast ptr addrspace(1) %output to ptr
%val = load i32, ptr %tmp0, align 4
store i32 %val, ptr %tmp1, align 4
ret void
}
; Optimized to group load/store.
define amdgpu_kernel void @load_store_group(ptr addrspace(3) nocapture %input, ptr addrspace(3) nocapture %output) #0 {
; CHECK-LABEL: define amdgpu_kernel void @load_store_group(
; CHECK-SAME: ptr addrspace(3) captures(none) [[INPUT:%.*]], ptr addrspace(3) captures(none) [[OUTPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[VAL:%.*]] = load i32, ptr addrspace(3) [[INPUT]], align 4
; CHECK-NEXT: store i32 [[VAL]], ptr addrspace(3) [[OUTPUT]], align 4
; CHECK-NEXT: ret void
;
%tmp0 = addrspacecast ptr addrspace(3) %input to ptr
%tmp1 = addrspacecast ptr addrspace(3) %output to ptr
%val = load i32, ptr %tmp0, align 4
store i32 %val, ptr %tmp1, align 4
ret void
}
; Optimized to private load/store.
define amdgpu_kernel void @load_store_private(ptr addrspace(5) nocapture %input, ptr addrspace(5) nocapture %output) #0 {
; CHECK-LABEL: define amdgpu_kernel void @load_store_private(
; CHECK-SAME: ptr addrspace(5) captures(none) [[INPUT:%.*]], ptr addrspace(5) captures(none) [[OUTPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[VAL:%.*]] = load i32, ptr addrspace(5) [[INPUT]], align 4
; CHECK-NEXT: store i32 [[VAL]], ptr addrspace(5) [[OUTPUT]], align 4
; CHECK-NEXT: ret void
;
%tmp0 = addrspacecast ptr addrspace(5) %input to ptr
%tmp1 = addrspacecast ptr addrspace(5) %output to ptr
%val = load i32, ptr %tmp0, align 4
store i32 %val, ptr %tmp1, align 4
ret void
}
; No optimization. flat load/store.
define amdgpu_kernel void @load_store_flat(ptr nocapture %input, ptr nocapture %output) #0 {
; CHECK-LABEL: define amdgpu_kernel void @load_store_flat(
; CHECK-SAME: ptr captures(none) [[INPUT:%.*]], ptr captures(none) [[OUTPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[INPUT]] to ptr addrspace(1)
; CHECK-NEXT: [[TMP2:%.*]] = addrspacecast ptr [[OUTPUT]] to ptr addrspace(1)
; CHECK-NEXT: [[VAL:%.*]] = load i32, ptr addrspace(1) [[TMP1]], align 4
; CHECK-NEXT: store i32 [[VAL]], ptr addrspace(1) [[TMP2]], align 4
; CHECK-NEXT: ret void
;
%val = load i32, ptr %input, align 4
store i32 %val, ptr %output, align 4
ret void
}
define amdgpu_kernel void @store_addrspacecast_ptr_value(ptr addrspace(1) nocapture %input, ptr addrspace(1) nocapture %output) #0 {
; CHECK-LABEL: define amdgpu_kernel void @store_addrspacecast_ptr_value(
; CHECK-SAME: ptr addrspace(1) captures(none) [[INPUT:%.*]], ptr addrspace(1) captures(none) [[OUTPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[CAST:%.*]] = addrspacecast ptr addrspace(1) [[INPUT]] to ptr
; CHECK-NEXT: store ptr [[CAST]], ptr addrspace(1) [[OUTPUT]], align 4
; CHECK-NEXT: ret void
;
%cast = addrspacecast ptr addrspace(1) %input to ptr
store ptr %cast, ptr addrspace(1) %output, align 4
ret void
}
define i32 @atomicrmw_add_global_to_flat(ptr addrspace(1) %global.ptr, i32 %y) #0 {
; CHECK-LABEL: define i32 @atomicrmw_add_global_to_flat(
; CHECK-SAME: ptr addrspace(1) [[GLOBAL_PTR:%.*]], i32 [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RET:%.*]] = atomicrmw add ptr addrspace(1) [[GLOBAL_PTR]], i32 [[Y]] seq_cst, align 4
; CHECK-NEXT: ret i32 [[RET]]
;
%cast = addrspacecast ptr addrspace(1) %global.ptr to ptr
%ret = atomicrmw add ptr %cast, i32 %y seq_cst
ret i32 %ret
}
define i32 @atomicrmw_add_group_to_flat(ptr addrspace(3) %group.ptr, i32 %y) #0 {
; CHECK-LABEL: define i32 @atomicrmw_add_group_to_flat(
; CHECK-SAME: ptr addrspace(3) [[GROUP_PTR:%.*]], i32 [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RET:%.*]] = atomicrmw add ptr addrspace(3) [[GROUP_PTR]], i32 [[Y]] seq_cst, align 4
; CHECK-NEXT: ret i32 [[RET]]
;
%cast = addrspacecast ptr addrspace(3) %group.ptr to ptr
%ret = atomicrmw add ptr %cast, i32 %y seq_cst
ret i32 %ret
}
define { i32, i1 } @cmpxchg_global_to_flat(ptr addrspace(1) %global.ptr, i32 %cmp, i32 %val) #0 {
; CHECK-LABEL: define { i32, i1 } @cmpxchg_global_to_flat(
; CHECK-SAME: ptr addrspace(1) [[GLOBAL_PTR:%.*]], i32 [[CMP:%.*]], i32 [[VAL:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RET:%.*]] = cmpxchg ptr addrspace(1) [[GLOBAL_PTR]], i32 [[CMP]], i32 [[VAL]] seq_cst monotonic, align 4
; CHECK-NEXT: ret { i32, i1 } [[RET]]
;
%cast = addrspacecast ptr addrspace(1) %global.ptr to ptr
%ret = cmpxchg ptr %cast, i32 %cmp, i32 %val seq_cst monotonic
ret { i32, i1 } %ret
}
define { i32, i1 } @cmpxchg_group_to_flat(ptr addrspace(3) %group.ptr, i32 %cmp, i32 %val) #0 {
; CHECK-LABEL: define { i32, i1 } @cmpxchg_group_to_flat(
; CHECK-SAME: ptr addrspace(3) [[GROUP_PTR:%.*]], i32 [[CMP:%.*]], i32 [[VAL:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RET:%.*]] = cmpxchg ptr addrspace(3) [[GROUP_PTR]], i32 [[CMP]], i32 [[VAL]] seq_cst monotonic, align 4
; CHECK-NEXT: ret { i32, i1 } [[RET]]
;
%cast = addrspacecast ptr addrspace(3) %group.ptr to ptr
%ret = cmpxchg ptr %cast, i32 %cmp, i32 %val seq_cst monotonic
ret { i32, i1 } %ret
}
; Not pointer operand
define { ptr, i1 } @cmpxchg_group_to_flat_wrong_operand(ptr addrspace(3) %cas.ptr, ptr addrspace(3) %cmp.ptr, ptr %val) #0 {
; CHECK-LABEL: define { ptr, i1 } @cmpxchg_group_to_flat_wrong_operand(
; CHECK-SAME: ptr addrspace(3) [[CAS_PTR:%.*]], ptr addrspace(3) [[CMP_PTR:%.*]], ptr [[VAL:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[CAST_CMP:%.*]] = addrspacecast ptr addrspace(3) [[CMP_PTR]] to ptr
; CHECK-NEXT: [[RET:%.*]] = cmpxchg ptr addrspace(3) [[CAS_PTR]], ptr [[CAST_CMP]], ptr [[VAL]] seq_cst monotonic, align 8
; CHECK-NEXT: ret { ptr, i1 } [[RET]]
;
%cast.cmp = addrspacecast ptr addrspace(3) %cmp.ptr to ptr
%ret = cmpxchg ptr addrspace(3) %cas.ptr, ptr %cast.cmp, ptr %val seq_cst monotonic
ret { ptr, i1 } %ret
}
; Null pointer in local addr space
define void @local_nullptr(ptr addrspace(1) nocapture %results, ptr addrspace(3) %a) {
; CHECK-LABEL: define void @local_nullptr(
; CHECK-SAME: ptr addrspace(1) captures(none) [[RESULTS:%.*]], ptr addrspace(3) [[A:%.*]]) {
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[TOBOOL:%.*]] = icmp ne ptr addrspace(3) [[A]], addrspacecast (ptr addrspace(5) null to ptr addrspace(3))
; CHECK-NEXT: [[CONV:%.*]] = zext i1 [[TOBOOL]] to i32
; CHECK-NEXT: store i32 [[CONV]], ptr addrspace(1) [[RESULTS]], align 4
; CHECK-NEXT: ret void
;
entry:
%tobool = icmp ne ptr addrspace(3) %a, addrspacecast (ptr addrspace(5) null to ptr addrspace(3))
%conv = zext i1 %tobool to i32
store i32 %conv, ptr addrspace(1) %results, align 4
ret void
}
define i32 @atomicrmw_add_global_to_flat_preserve_amdgpu_md(ptr addrspace(1) %global.ptr, i32 %y) #0 {
; CHECK-LABEL: define i32 @atomicrmw_add_global_to_flat_preserve_amdgpu_md(
; CHECK-SAME: ptr addrspace(1) [[GLOBAL_PTR:%.*]], i32 [[Y:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[RET:%.*]] = atomicrmw add ptr addrspace(1) [[GLOBAL_PTR]], i32 [[Y]] seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META0:![0-9]+]], !amdgpu.no.remote.memory [[META0]]
; CHECK-NEXT: ret i32 [[RET]]
;
%cast = addrspacecast ptr addrspace(1) %global.ptr to ptr
%ret = atomicrmw add ptr %cast, i32 %y seq_cst, align 4, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
ret i32 %ret
}
; Make sure there's no assert
define ptr @try_infer_getelementptr_constant_null() {
; CHECK-LABEL: define ptr @try_infer_getelementptr_constant_null() {
; CHECK-NEXT: [[CE:%.*]] = getelementptr i8, ptr getelementptr inbounds (i8, ptr null, i64 8), i64 0
; CHECK-NEXT: ret ptr [[CE]]
;
%ce = getelementptr i8, ptr getelementptr inbounds (i8, ptr null, i64 8), i64 0
ret ptr %ce
}
attributes #0 = { nounwind }
!0 = !{}
;.
; CHECK: [[META0]] = !{}
;.
|
