1 ; RUN: llc -march=amdgcn -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,SI,SICIVI %s
2 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,CI,SICIVI,CIPLUS %s
3 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI,SICIVI,CIPLUS %s
4 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-flat-for-global -verify-machineinstrs< %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9,CIPLUS %s
6 ; GCN-LABEL: {{^}}local_i32_load
10 ; GCN: ds_read_b32 [[REG:v[0-9]+]], v{{[0-9]+}} offset:28
11 ; GCN: buffer_store_dword [[REG]],
12 define amdgpu_kernel void @local_i32_load(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
13 %gep = getelementptr i32, ptr addrspace(3) %in, i32 7
14 %val = load i32, ptr addrspace(3) %gep, align 4
15 store i32 %val, ptr addrspace(1) %out, align 4
19 ; GCN-LABEL: {{^}}local_i32_load_0_offset
20 ; SICIVI: s_mov_b32 m0
23 ; GCN: ds_read_b32 [[REG:v[0-9]+]], v{{[0-9]+}}
24 ; GCN: buffer_store_dword [[REG]],
25 define amdgpu_kernel void @local_i32_load_0_offset(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
26 %val = load i32, ptr addrspace(3) %in, align 4
27 store i32 %val, ptr addrspace(1) %out, align 4
31 ; GCN-LABEL: {{^}}local_i8_load_i16_max_offset:
32 ; SICIVI: s_mov_b32 m0
36 ; GCN: ds_read_u8 [[REG:v[0-9]+]], {{v[0-9]+}} offset:65535
37 ; GCN: buffer_store_byte [[REG]],
38 define amdgpu_kernel void @local_i8_load_i16_max_offset(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
39 %gep = getelementptr i8, ptr addrspace(3) %in, i32 65535
40 %val = load i8, ptr addrspace(3) %gep, align 4
41 store i8 %val, ptr addrspace(1) %out, align 4
45 ; GCN-LABEL: {{^}}local_i8_load_over_i16_max_offset:
46 ; SICIVI-DAG: s_mov_b32 m0
49 ; The LDS offset will be 65536 bytes, which is larger than the size of LDS on
50 ; SI, which is why it is being OR'd with the base pointer.
51 ; SI-DAG: s_bitset1_b32 [[ADDR:s[0-9]+]], 16
52 ; CI-DAG: s_add_i32 [[ADDR:s[0-9]+]], s{{[0-9]+}}, 0x10000
53 ; VI-DAG: s_add_i32 [[ADDR:s[0-9]+]], s{{[0-9]+}}, 0x10000
54 ; GFX9-DAG: s_add_i32 [[ADDR:s[0-9]+]], s{{[0-9]+}}, 0x10000
56 ; GCN-DAG: v_mov_b32_e32 [[VREGADDR:v[0-9]+]], [[ADDR]]
57 ; GCN: ds_read_u8 [[REG:v[0-9]+]], [[VREGADDR]]
58 ; GCN: buffer_store_byte [[REG]],
59 define amdgpu_kernel void @local_i8_load_over_i16_max_offset(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
60 %gep = getelementptr i8, ptr addrspace(3) %in, i32 65536
61 %val = load i8, ptr addrspace(3) %gep, align 4
62 store i8 %val, ptr addrspace(1) %out, align 4
66 ; GCN-LABEL: {{^}}local_i64_load:
67 ; SICIVI: s_mov_b32 m0
71 ; GCN: ds_read_b64 [[REG:v[[0-9]+:[0-9]+]]], v{{[0-9]+}} offset:56
72 ; GCN: buffer_store_dwordx2 [[REG]],
73 define amdgpu_kernel void @local_i64_load(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
74 %gep = getelementptr i64, ptr addrspace(3) %in, i32 7
75 %val = load i64, ptr addrspace(3) %gep, align 8
76 store i64 %val, ptr addrspace(1) %out, align 8
80 ; GCN-LABEL: {{^}}local_i64_load_0_offset
81 ; SICIVI: s_mov_b32 m0
84 ; GCN: ds_read_b64 [[REG:v\[[0-9]+:[0-9]+\]]], v{{[0-9]+}}
85 ; GCN: buffer_store_dwordx2 [[REG]],
86 define amdgpu_kernel void @local_i64_load_0_offset(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
87 %val = load i64, ptr addrspace(3) %in, align 8
88 store i64 %val, ptr addrspace(1) %out, align 8
92 ; GCN-LABEL: {{^}}local_f64_load:
93 ; SICIVI: s_mov_b32 m0
97 ; GCN: ds_read_b64 [[REG:v[[0-9]+:[0-9]+]]], v{{[0-9]+}} offset:56
98 ; GCN: buffer_store_dwordx2 [[REG]],
99 define amdgpu_kernel void @local_f64_load(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
100 %gep = getelementptr double, ptr addrspace(3) %in, i32 7
101 %val = load double, ptr addrspace(3) %gep, align 8
102 store double %val, ptr addrspace(1) %out, align 8
106 ; GCN-LABEL: {{^}}local_f64_load_0_offset
107 ; SICIVI: s_mov_b32 m0
110 ; GCN: ds_read_b64 [[REG:v\[[0-9]+:[0-9]+\]]], v{{[0-9]+}}
111 ; GCN: buffer_store_dwordx2 [[REG]],
112 define amdgpu_kernel void @local_f64_load_0_offset(ptr addrspace(1) %out, ptr addrspace(3) %in) nounwind {
113 %val = load double, ptr addrspace(3) %in, align 8
114 store double %val, ptr addrspace(1) %out, align 8
118 ; GCN-LABEL: {{^}}local_i64_store:
119 ; SICIVI: s_mov_b32 m0
123 ; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:56
124 define amdgpu_kernel void @local_i64_store(ptr addrspace(3) %out) nounwind {
125 %gep = getelementptr i64, ptr addrspace(3) %out, i32 7
126 store i64 5678, ptr addrspace(3) %gep, align 8
130 ; GCN-LABEL: {{^}}local_i64_store_0_offset:
131 ; SICIVI: s_mov_b32 m0
135 ; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}
136 define amdgpu_kernel void @local_i64_store_0_offset(ptr addrspace(3) %out) nounwind {
137 store i64 1234, ptr addrspace(3) %out, align 8
141 ; GCN-LABEL: {{^}}local_f64_store:
142 ; SICIVI: s_mov_b32 m0
146 ; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:56
147 define amdgpu_kernel void @local_f64_store(ptr addrspace(3) %out) nounwind {
148 %gep = getelementptr double, ptr addrspace(3) %out, i32 7
149 store double 16.0, ptr addrspace(3) %gep, align 8
153 ; GCN-LABEL: {{^}}local_f64_store_0_offset
154 ; SICIVI: s_mov_b32 m0
157 ; GCN: ds_write_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}
158 define amdgpu_kernel void @local_f64_store_0_offset(ptr addrspace(3) %out) nounwind {
159 store double 20.0, ptr addrspace(3) %out, align 8
163 ; GCN-LABEL: {{^}}local_v2i64_store:
164 ; SICIVI: s_mov_b32 m0
168 ; SI: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:14 offset1:15
169 ; CIPLUS: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:112
171 define amdgpu_kernel void @local_v2i64_store(ptr addrspace(3) %out) nounwind {
172 %gep = getelementptr <2 x i64>, ptr addrspace(3) %out, i32 7
173 store <2 x i64> <i64 5678, i64 5678>, ptr addrspace(3) %gep, align 16
177 ; GCN-LABEL: {{^}}local_v2i64_store_0_offset:
178 ; SICIVI: s_mov_b32 m0
183 ; SI: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1{{$}}
184 ; CIPLUS: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]$}}
187 define amdgpu_kernel void @local_v2i64_store_0_offset(ptr addrspace(3) %out) nounwind {
188 store <2 x i64> <i64 1234, i64 1234>, ptr addrspace(3) %out, align 16
192 ; GCN-LABEL: {{^}}local_v4i64_store:
193 ; SICIVI: s_mov_b32 m0
197 ; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:30 offset1:31
198 ; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:28 offset1:29
200 ; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:224{{$}}
201 ; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:240{{$}}
204 define amdgpu_kernel void @local_v4i64_store(ptr addrspace(3) %out) nounwind {
205 %gep = getelementptr <4 x i64>, ptr addrspace(3) %out, i32 7
206 store <4 x i64> <i64 5678, i64 5678, i64 5678, i64 5678>, ptr addrspace(3) %gep, align 16
210 ; GCN-LABEL: {{^}}local_v4i64_store_0_offset:
211 ; SICIVI: s_mov_b32 m0
215 ; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset0:2 offset1:3
216 ; SI-DAG: ds_write2_b64 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}} offset1:1
218 ; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]$}}
219 ; CIPLUS-DAG: ds_write_b128 v{{[0-9]+}}, {{v\[[0-9]+:[0-9]+\]}} offset:16{{$}}
222 define amdgpu_kernel void @local_v4i64_store_0_offset(ptr addrspace(3) %out) nounwind {
223 store <4 x i64> <i64 1234, i64 1234, i64 1234, i64 1234>, ptr addrspace(3) %out, align 16