1; RUN: opt -S -mtriple=amdgcn-unknown-unknown -amdgpu-annotate-kernel-features < %s | FileCheck -check-prefix=NOHSA -check-prefix=ALL %s
2
3declare i32 @llvm.r600.read.tgid.x() #0
4declare i32 @llvm.r600.read.tgid.y() #0
5declare i32 @llvm.r600.read.tgid.z() #0
6
7declare i32 @llvm.r600.read.tidig.x() #0
8declare i32 @llvm.r600.read.tidig.y() #0
9declare i32 @llvm.r600.read.tidig.z() #0
10
11declare i32 @llvm.r600.read.local.size.x() #0
12declare i32 @llvm.r600.read.local.size.y() #0
13declare i32 @llvm.r600.read.local.size.z() #0
14
15; ALL: define amdgpu_kernel void @use_tgid_x(i32 addrspace(1)* %ptr) #1 {
16define amdgpu_kernel void @use_tgid_x(i32 addrspace(1)* %ptr) #1 {
17  %val = call i32 @llvm.r600.read.tgid.x()
18  store i32 %val, i32 addrspace(1)* %ptr
19  ret void
20}
21
22; ALL: define amdgpu_kernel void @use_tgid_y(i32 addrspace(1)* %ptr) #2 {
23define amdgpu_kernel void @use_tgid_y(i32 addrspace(1)* %ptr) #1 {
24  %val = call i32 @llvm.r600.read.tgid.y()
25  store i32 %val, i32 addrspace(1)* %ptr
26  ret void
27}
28
29; ALL: define amdgpu_kernel void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #2 {
30define amdgpu_kernel void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #1 {
31  %val0 = call i32 @llvm.r600.read.tgid.y()
32  store volatile i32 %val0, i32 addrspace(1)* %ptr
33  %val1 = call i32 @llvm.r600.read.tgid.y()
34  store volatile i32 %val1, i32 addrspace(1)* %ptr
35  ret void
36}
37
38; ALL: define amdgpu_kernel void @use_tgid_x_y(i32 addrspace(1)* %ptr) #2 {
39define amdgpu_kernel void @use_tgid_x_y(i32 addrspace(1)* %ptr) #1 {
40  %val0 = call i32 @llvm.r600.read.tgid.x()
41  %val1 = call i32 @llvm.r600.read.tgid.y()
42  store volatile i32 %val0, i32 addrspace(1)* %ptr
43  store volatile i32 %val1, i32 addrspace(1)* %ptr
44  ret void
45}
46
47; ALL: define amdgpu_kernel void @use_tgid_z(i32 addrspace(1)* %ptr) #3 {
48define amdgpu_kernel void @use_tgid_z(i32 addrspace(1)* %ptr) #1 {
49  %val = call i32 @llvm.r600.read.tgid.z()
50  store i32 %val, i32 addrspace(1)* %ptr
51  ret void
52}
53
54; ALL: define amdgpu_kernel void @use_tgid_x_z(i32 addrspace(1)* %ptr) #3 {
55define amdgpu_kernel void @use_tgid_x_z(i32 addrspace(1)* %ptr) #1 {
56  %val0 = call i32 @llvm.r600.read.tgid.x()
57  %val1 = call i32 @llvm.r600.read.tgid.z()
58  store volatile i32 %val0, i32 addrspace(1)* %ptr
59  store volatile i32 %val1, i32 addrspace(1)* %ptr
60  ret void
61}
62
63; ALL: define amdgpu_kernel void @use_tgid_y_z(i32 addrspace(1)* %ptr) #4 {
64define amdgpu_kernel void @use_tgid_y_z(i32 addrspace(1)* %ptr) #1 {
65  %val0 = call i32 @llvm.r600.read.tgid.y()
66  %val1 = call i32 @llvm.r600.read.tgid.z()
67  store volatile i32 %val0, i32 addrspace(1)* %ptr
68  store volatile i32 %val1, i32 addrspace(1)* %ptr
69  ret void
70}
71
72; ALL: define amdgpu_kernel void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #4 {
73define amdgpu_kernel void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #1 {
74  %val0 = call i32 @llvm.r600.read.tgid.x()
75  %val1 = call i32 @llvm.r600.read.tgid.y()
76  %val2 = call i32 @llvm.r600.read.tgid.z()
77  store volatile i32 %val0, i32 addrspace(1)* %ptr
78  store volatile i32 %val1, i32 addrspace(1)* %ptr
79  store volatile i32 %val2, i32 addrspace(1)* %ptr
80  ret void
81}
82
83; ALL: define amdgpu_kernel void @use_tidig_x(i32 addrspace(1)* %ptr) #1 {
84define amdgpu_kernel void @use_tidig_x(i32 addrspace(1)* %ptr) #1 {
85  %val = call i32 @llvm.r600.read.tidig.x()
86  store i32 %val, i32 addrspace(1)* %ptr
87  ret void
88}
89
90; ALL: define amdgpu_kernel void @use_tidig_y(i32 addrspace(1)* %ptr) #5 {
91define amdgpu_kernel void @use_tidig_y(i32 addrspace(1)* %ptr) #1 {
92  %val = call i32 @llvm.r600.read.tidig.y()
93  store i32 %val, i32 addrspace(1)* %ptr
94  ret void
95}
96
97; ALL: define amdgpu_kernel void @use_tidig_z(i32 addrspace(1)* %ptr) #6 {
98define amdgpu_kernel void @use_tidig_z(i32 addrspace(1)* %ptr) #1 {
99  %val = call i32 @llvm.r600.read.tidig.z()
100  store i32 %val, i32 addrspace(1)* %ptr
101  ret void
102}
103
104; ALL: define amdgpu_kernel void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
105define amdgpu_kernel void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
106  %val0 = call i32 @llvm.r600.read.tidig.x()
107  %val1 = call i32 @llvm.r600.read.tgid.x()
108  store volatile i32 %val0, i32 addrspace(1)* %ptr
109  store volatile i32 %val1, i32 addrspace(1)* %ptr
110  ret void
111}
112
113; ALL: define amdgpu_kernel void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #7 {
114define amdgpu_kernel void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #1 {
115  %val0 = call i32 @llvm.r600.read.tidig.y()
116  %val1 = call i32 @llvm.r600.read.tgid.y()
117  store volatile i32 %val0, i32 addrspace(1)* %ptr
118  store volatile i32 %val1, i32 addrspace(1)* %ptr
119  ret void
120}
121
122; ALL: define amdgpu_kernel void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #8 {
123define amdgpu_kernel void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #1 {
124  %val0 = call i32 @llvm.r600.read.tidig.x()
125  %val1 = call i32 @llvm.r600.read.tidig.y()
126  %val2 = call i32 @llvm.r600.read.tidig.z()
127  store volatile i32 %val0, i32 addrspace(1)* %ptr
128  store volatile i32 %val1, i32 addrspace(1)* %ptr
129  store volatile i32 %val2, i32 addrspace(1)* %ptr
130  ret void
131}
132
133; ALL: define amdgpu_kernel void @use_all_workitems(i32 addrspace(1)* %ptr) #9 {
134define amdgpu_kernel void @use_all_workitems(i32 addrspace(1)* %ptr) #1 {
135  %val0 = call i32 @llvm.r600.read.tidig.x()
136  %val1 = call i32 @llvm.r600.read.tidig.y()
137  %val2 = call i32 @llvm.r600.read.tidig.z()
138  %val3 = call i32 @llvm.r600.read.tgid.x()
139  %val4 = call i32 @llvm.r600.read.tgid.y()
140  %val5 = call i32 @llvm.r600.read.tgid.z()
141  store volatile i32 %val0, i32 addrspace(1)* %ptr
142  store volatile i32 %val1, i32 addrspace(1)* %ptr
143  store volatile i32 %val2, i32 addrspace(1)* %ptr
144  store volatile i32 %val3, i32 addrspace(1)* %ptr
145  store volatile i32 %val4, i32 addrspace(1)* %ptr
146  store volatile i32 %val5, i32 addrspace(1)* %ptr
147  ret void
148}
149
150; HSA: define amdgpu_kernel void @use_get_local_size_x(i32 addrspace(1)* %ptr) #10 {
151; NOHSA: define amdgpu_kernel void @use_get_local_size_x(i32 addrspace(1)* %ptr) #1 {
152define amdgpu_kernel void @use_get_local_size_x(i32 addrspace(1)* %ptr) #1 {
153  %val = call i32 @llvm.r600.read.local.size.x()
154  store i32 %val, i32 addrspace(1)* %ptr
155  ret void
156}
157
158; HSA: define amdgpu_kernel void @use_get_local_size_y(i32 addrspace(1)* %ptr) #10 {
159; NOHSA: define amdgpu_kernel void @use_get_local_size_y(i32 addrspace(1)* %ptr) #1 {
160define amdgpu_kernel void @use_get_local_size_y(i32 addrspace(1)* %ptr) #1 {
161  %val = call i32 @llvm.r600.read.local.size.y()
162  store i32 %val, i32 addrspace(1)* %ptr
163  ret void
164}
165
166; HSA: define amdgpu_kernel void @use_get_local_size_z(i32 addrspace(1)* %ptr) #10 {
167; NOHSA: define amdgpu_kernel void @use_get_local_size_z(i32 addrspace(1)* %ptr) #1 {
168define amdgpu_kernel void @use_get_local_size_z(i32 addrspace(1)* %ptr) #1 {
169  %val = call i32 @llvm.r600.read.local.size.z()
170  store i32 %val, i32 addrspace(1)* %ptr
171  ret void
172}
173
174attributes #0 = { nounwind readnone }
175attributes #1 = { nounwind }
176
177; HSA: attributes #0 = { nounwind readnone }
178; HSA: attributes #1 = { nounwind }
179; HSA: attributes #2 = { nounwind "amdgpu-work-group-id-y" }
180; HSA: attributes #3 = { nounwind "amdgpu-work-group-id-z" }
181; HSA: attributes #4 = { nounwind "amdgpu-work-group-id-y" "amdgpu-work-group-id-z" }
182; HSA: attributes #5 = { nounwind "amdgpu-work-item-id-y" }
183; HSA: attributes #6 = { nounwind "amdgpu-work-item-id-z" }
184; HSA: attributes #7 = { nounwind "amdgpu-work-group-id-y" "amdgpu-work-item-id-y" }
185; HSA: attributes #8 = { nounwind "amdgpu-work-item-id-y" "amdgpu-work-item-id-z" }
186; HSA: attributes #9 = { nounwind "amdgpu-work-group-id-y" "amdgpu-work-group-id-z" "amdgpu-work-item-id-y" "amdgpu-work-item-id-z" }
187; HSA: attributes #10 = { nounwind "amdgpu-dispatch-ptr" }
188