1; RUN: llc -march=r600 -mcpu=juniper < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
2
3; === WIDTH ==================================================================
4; 9 implicit args = 9 dwords to first image argument.
5; First width at dword index 9+1 -> KC0[2].Z
6
7; FUNC-LABEL: {{^}}width_2d:
8; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
9; EG: MOV * [[VAL]], KC0[2].Z
10define amdgpu_kernel void @width_2d (%opencl.image2d_t addrspace(1)* %in,
11                       i32 addrspace(1)* %out) {
12entry:
13  %0 = call [3 x i32] @llvm.OpenCL.image.get.size.2d(
14      %opencl.image2d_t addrspace(1)* %in) #0
15  %1 = extractvalue [3 x i32] %0, 0
16  store i32 %1, i32 addrspace(1)* %out
17  ret void
18}
19
20; FUNC-LABEL: {{^}}width_3d:
21; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
22; EG: MOV * [[VAL]], KC0[2].Z
23define amdgpu_kernel void @width_3d (%opencl.image3d_t addrspace(1)* %in,
24                       i32 addrspace(1)* %out) {
25entry:
26  %0 = call [3 x i32] @llvm.OpenCL.image.get.size.3d(
27      %opencl.image3d_t addrspace(1)* %in) #0
28  %1 = extractvalue [3 x i32] %0, 0
29  store i32 %1, i32 addrspace(1)* %out
30  ret void
31}
32
33
34; === HEIGHT =================================================================
35; First height at dword index 9+2 -> KC0[2].W
36
37; FUNC-LABEL: {{^}}height_2d:
38; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
39; EG: MOV * [[VAL]], KC0[2].W
40define amdgpu_kernel void @height_2d (%opencl.image2d_t addrspace(1)* %in,
41                        i32 addrspace(1)* %out) {
42entry:
43  %0 = call [3 x i32] @llvm.OpenCL.image.get.size.2d(
44      %opencl.image2d_t addrspace(1)* %in) #0
45  %1 = extractvalue [3 x i32] %0, 1
46  store i32 %1, i32 addrspace(1)* %out
47  ret void
48}
49
50; FUNC-LABEL: {{^}}height_3d:
51; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
52; EG: MOV * [[VAL]], KC0[2].W
53define amdgpu_kernel void @height_3d (%opencl.image3d_t addrspace(1)* %in,
54                        i32 addrspace(1)* %out) {
55entry:
56  %0 = call [3 x i32] @llvm.OpenCL.image.get.size.3d(
57      %opencl.image3d_t addrspace(1)* %in) #0
58  %1 = extractvalue [3 x i32] %0, 1
59  store i32 %1, i32 addrspace(1)* %out
60  ret void
61}
62
63
64; === DEPTH ==================================================================
65; First depth at dword index 9+3 -> KC0[3].X
66
67; FUNC-LABEL: {{^}}depth_3d:
68; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
69; EG: MOV * [[VAL]], KC0[3].X
70define amdgpu_kernel void @depth_3d (%opencl.image3d_t addrspace(1)* %in,
71                       i32 addrspace(1)* %out) {
72entry:
73  %0 = call [3 x i32] @llvm.OpenCL.image.get.size.3d(
74      %opencl.image3d_t addrspace(1)* %in) #0
75  %1 = extractvalue [3 x i32] %0, 2
76  store i32 %1, i32 addrspace(1)* %out
77  ret void
78}
79
80
81; === CHANNEL DATA TYPE ======================================================
82; First channel data type at dword index 9+4 -> KC0[3].Y
83
84; FUNC-LABEL: {{^}}data_type_2d:
85; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
86; EG: MOV * [[VAL]], KC0[3].Y
87define amdgpu_kernel void @data_type_2d (%opencl.image2d_t addrspace(1)* %in,
88                           i32 addrspace(1)* %out) {
89entry:
90  %0 = call [2 x i32] @llvm.OpenCL.image.get.format.2d(
91      %opencl.image2d_t addrspace(1)* %in) #0
92  %1 = extractvalue [2 x i32] %0, 0
93  store i32 %1, i32 addrspace(1)* %out
94  ret void
95}
96
97; FUNC-LABEL: {{^}}data_type_3d:
98; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
99; EG: MOV * [[VAL]], KC0[3].Y
100define amdgpu_kernel void @data_type_3d (%opencl.image3d_t addrspace(1)* %in,
101                                     i32 addrspace(1)* %out) {
102entry:
103  %0 = call [2 x i32] @llvm.OpenCL.image.get.format.3d(
104      %opencl.image3d_t addrspace(1)* %in) #0
105  %1 = extractvalue [2 x i32] %0, 0
106  store i32 %1, i32 addrspace(1)* %out
107  ret void
108}
109
110
111; === CHANNEL ORDER ==========================================================
112; First channel order at dword index 9+5 -> KC0[3].Z
113
114; FUNC-LABEL: {{^}}channel_order_2d:
115; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
116; EG: MOV * [[VAL]], KC0[3].Z
117define amdgpu_kernel void @channel_order_2d (%opencl.image2d_t addrspace(1)* %in,
118                               i32 addrspace(1)* %out) {
119entry:
120  %0 = call [2 x i32] @llvm.OpenCL.image.get.format.2d(
121      %opencl.image2d_t addrspace(1)* %in) #0
122  %1 = extractvalue [2 x i32] %0, 1
123  store i32 %1, i32 addrspace(1)* %out
124  ret void
125}
126
127; FUNC-LABEL: {{^}}channel_order_3d:
128; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
129; EG: MOV * [[VAL]], KC0[3].Z
130define amdgpu_kernel void @channel_order_3d (%opencl.image3d_t addrspace(1)* %in,
131                                         i32 addrspace(1)* %out) {
132entry:
133  %0 = call [2 x i32] @llvm.OpenCL.image.get.format.3d(
134      %opencl.image3d_t addrspace(1)* %in) #0
135  %1 = extractvalue [2 x i32] %0, 1
136  store i32 %1, i32 addrspace(1)* %out
137  ret void
138}
139
140
141; === 2ND IMAGE ==============================================================
142; 9 implicit args + 2 explicit args + 5 implicit args for 1st image argument
143;   = 16 dwords to 2nd image argument.
144; Height of the second image is at 16+2 -> KC0[4].Z
145;
146; FUNC-LABEL: {{^}}image_arg_2nd:
147; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
148; EG: MOV * [[VAL]], KC0[4].Z
149define amdgpu_kernel void @image_arg_2nd (%opencl.image3d_t addrspace(1)* %in1,
150                            i32 %x,
151                            %opencl.image2d_t addrspace(1)* %in2,
152                            i32 addrspace(1)* %out) {
153entry:
154  %0 = call [3 x i32] @llvm.OpenCL.image.get.size.2d(
155      %opencl.image2d_t addrspace(1)* %in2) #0
156  %1 = extractvalue [3 x i32] %0, 1
157  store i32 %1, i32 addrspace(1)* %out
158  ret void
159}
160
161%opencl.image2d_t = type opaque
162%opencl.image3d_t = type opaque
163
164declare [3 x i32] @llvm.OpenCL.image.get.size.2d(%opencl.image2d_t addrspace(1)*) #0
165declare [3 x i32] @llvm.OpenCL.image.get.size.3d(%opencl.image3d_t addrspace(1)*) #0
166declare [2 x i32] @llvm.OpenCL.image.get.format.2d(%opencl.image2d_t addrspace(1)*) #0
167declare [2 x i32] @llvm.OpenCL.image.get.format.3d(%opencl.image3d_t addrspace(1)*) #0
168
169attributes #0 = { readnone }
170
171!opencl.kernels = !{!0, !1, !2, !3, !4, !5, !6, !7, !8, !9}
172!0 = !{void (%opencl.image2d_t addrspace(1)*, i32 addrspace(1)*)* @width_2d,
173       !10, !20, !30, !40, !50}
174!1 = !{void (%opencl.image3d_t addrspace(1)*, i32 addrspace(1)*)* @width_3d,
175       !10, !21, !31, !41, !50}
176!2 = !{void (%opencl.image2d_t addrspace(1)*, i32 addrspace(1)*)* @height_2d,
177       !10, !20, !30, !40, !50}
178!3 = !{void (%opencl.image3d_t addrspace(1)*, i32 addrspace(1)*)* @height_3d,
179       !10, !21, !31, !41, !50}
180!4 = !{void (%opencl.image3d_t addrspace(1)*, i32 addrspace(1)*)* @depth_3d,
181       !10, !21, !31, !41, !50}
182!5 = !{void (%opencl.image2d_t addrspace(1)*, i32 addrspace(1)*)* @data_type_2d,
183       !10, !20, !30, !40, !50}
184!6 = !{void (%opencl.image3d_t addrspace(1)*, i32 addrspace(1)*)* @data_type_3d,
185       !10, !21, !31, !41, !50}
186!7 = !{void (%opencl.image2d_t addrspace(1)*, i32 addrspace(1)*)* @channel_order_2d,
187       !10, !20, !30, !40, !50}
188!8 = !{void (%opencl.image3d_t addrspace(1)*, i32 addrspace(1)*)* @channel_order_3d,
189       !10, !21, !31, !41, !50}
190!9 = !{void (%opencl.image3d_t addrspace(1)*, i32, %opencl.image2d_t addrspace(1)*,
191      i32 addrspace(1)*)* @image_arg_2nd, !12, !22, !32, !42, !52}
192
193!10 = !{!"kernel_arg_addr_space", i32 1, i32 1}
194!20 = !{!"kernel_arg_access_qual", !"read_only", !"none"}
195!21 = !{!"kernel_arg_access_qual", !"read_only", !"none"}
196!30 = !{!"kernel_arg_type", !"image2d_t", !"int*"}
197!31 = !{!"kernel_arg_type", !"image3d_t", !"int*"}
198!40 = !{!"kernel_arg_base_type", !"image2d_t", !"int*"}
199!41 = !{!"kernel_arg_base_type", !"image3d_t", !"int*"}
200!50 = !{!"kernel_arg_type_qual", !"", !""}
201
202!12 = !{!"kernel_arg_addr_space", i32 1, i32 0, i32 1, i32 1}
203!22 = !{!"kernel_arg_access_qual", !"read_only", !"none", !"write_only", !"none"}
204!32 = !{!"kernel_arg_type", !"image3d_t", !"sampler_t", !"image2d_t", !"int*"}
205!42 = !{!"kernel_arg_base_type", !"image3d_t", !"sampler_t", !"image2d_t", !"int*"}
206!52 = !{!"kernel_arg_type_qual", !"", !"", !"", !""}
207