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
|
; RUN: opt -S -disable-promote-alloca-to-vector -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=amdgpu-promote-alloca < %s | FileCheck -check-prefix=IR %s
; RUN: llc -disable-promote-alloca-to-vector -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -amdgpu-enable-lower-module-lds=false < %s | FileCheck -check-prefix=ASM %s
@all_lds = internal unnamed_addr addrspace(3) global [16384 x i32] poison, align 4
@some_lds = internal unnamed_addr addrspace(3) global [32 x i32] poison, align 4
@some_dynamic_lds = external hidden addrspace(3) global [0 x i32], align 4
@initializer_user_some = addrspace(1) global i32 ptrtoint (ptr addrspace(3) @some_lds to i32), align 4
@initializer_user_all = addrspace(1) global i32 ptrtoint (ptr addrspace(3) @all_lds to i32), align 4
; This function cannot promote to using LDS because of the size of the
; constant expression use in the function, which was previously not
; detected.
; IR-LABEL: @constant_expression_uses_all_lds(
; IR: alloca
; ASM-LABEL: constant_expression_uses_all_lds:
; ASM: .amdhsa_group_segment_fixed_size 65536
define amdgpu_kernel void @constant_expression_uses_all_lds(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
store volatile i32 ptrtoint (ptr addrspace(3) @all_lds to i32), ptr addrspace(1) poison
ret void
}
; Has a constant expression use through a single level of constant
; expression, but not enough LDS to block promotion
; IR-LABEL: @constant_expression_uses_some_lds(
; IR-NOT: alloca
; ASM-LABEL: {{^}}constant_expression_uses_some_lds:
; ASM: .amdhsa_group_segment_fixed_size 4224{{$}}
define amdgpu_kernel void @constant_expression_uses_some_lds(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
store volatile i32 ptrtoint (ptr addrspace(3) @some_lds to i32), ptr addrspace(1) poison
ret void
}
; Has a constant expression use through a single level of constant
; expression, but usage of dynamic LDS should block promotion
; IR-LABEL: @constant_expression_uses_some_dynamic_lds(
; IR: alloca
; ASM-LABEL: {{^}}constant_expression_uses_some_dynamic_lds:
; ASM: .amdhsa_group_segment_fixed_size 0{{$}}
define amdgpu_kernel void @constant_expression_uses_some_dynamic_lds(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
store i32 1234, ptr addrspacecast (ptr addrspace(3) @some_dynamic_lds to ptr), align 4
ret void
}
declare void @callee(ptr)
; IR-LABEL: @constant_expression_uses_all_lds_multi_level(
; IR: alloca
; ASM-LABEL: {{^}}constant_expression_uses_all_lds_multi_level:
; ASM: .amdhsa_group_segment_fixed_size 65536{{$}}
define amdgpu_kernel void @constant_expression_uses_all_lds_multi_level(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
call void @callee(ptr addrspacecast (ptr addrspace(3) getelementptr inbounds ([16384 x i32], ptr addrspace(3) @all_lds, i32 0, i32 8) to ptr))
ret void
}
; IR-LABEL: @constant_expression_uses_some_lds_multi_level(
; IR-NOT: alloca
; IR: llvm.amdgcn.workitem.id
; ASM-LABEL: {{^}}constant_expression_uses_some_lds_multi_level:
; ASM: .amdhsa_group_segment_fixed_size 4224{{$}}
define amdgpu_kernel void @constant_expression_uses_some_lds_multi_level(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
call void @callee(ptr addrspacecast (ptr addrspace(3) getelementptr inbounds ([32 x i32], ptr addrspace(3) @some_lds, i32 0, i32 8) to ptr))
ret void
}
; IR-LABEL: @constant_expression_uses_some_dynamic_lds_multi_level(
; IR: alloca
; ASM-LABEL: {{^}}constant_expression_uses_some_dynamic_lds_multi_level:
; ASM: .amdhsa_group_segment_fixed_size 0{{$}}
define amdgpu_kernel void @constant_expression_uses_some_dynamic_lds_multi_level(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
call void @callee(ptr addrspacecast (ptr addrspace(3) @some_dynamic_lds to ptr))
ret void
}
; IR-LABEL: @constant_expression_uses_some_lds_global_initializer(
; IR-NOT: alloca
; IR: llvm.amdgcn.workitem.id
; ASM-LABEL: {{^}}constant_expression_uses_some_lds_global_initializer:
; ASM: .amdhsa_group_segment_fixed_size 4096{{$}}
define amdgpu_kernel void @constant_expression_uses_some_lds_global_initializer(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
store volatile i32 ptrtoint (ptr addrspace(1) @initializer_user_some to i32), ptr addrspace(1) poison
ret void
}
; We can't actually handle LDS initializers in global initializers,
; but this should count as usage.
; IR-LABEL: @constant_expression_uses_all_lds_global_initializer(
; IR: alloca
; ASM-LABEL: {{^}}constant_expression_uses_all_lds_global_initializer:
; ASM: .group_segment_fixed_size: 65536
define amdgpu_kernel void @constant_expression_uses_all_lds_global_initializer(ptr addrspace(1) nocapture %out, i32 %idx) #0 {
entry:
%stack = alloca [4 x i32], align 4, addrspace(5)
%gep1 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 1
%gep2 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 2
%gep3 = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 3
store i32 9, ptr addrspace(5) %stack
store i32 10, ptr addrspace(5) %gep1
store i32 99, ptr addrspace(5) %gep2
store i32 43, ptr addrspace(5) %gep3
%arrayidx = getelementptr inbounds [4 x i32], ptr addrspace(5) %stack, i32 0, i32 %idx
%load = load i32, ptr addrspace(5) %arrayidx, align 4
store i32 %load, ptr addrspace(1) %out
store volatile i32 ptrtoint (ptr addrspace(1) @initializer_user_all to i32), ptr addrspace(1) poison
ret void
}
attributes #0 = { "amdgpu-waves-per-eu"="1,5" "amdgpu-flat-work-group-size"="256,256" }
|
