1 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs -mattr=+load-store-opt < %s | FileCheck -strict-whitespace -check-prefix=SI %s
3 @lds = addrspace(3) global [512 x float] undef, align 4
4 @lds.f64 = addrspace(3) global [512 x double] undef, align 8
7 ; SI-LABEL: @simple_write2_one_val_f32
8 ; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]]
9 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 2, v{{[0-9]+}}
10 ; SI: ds_write2_b32 [[VPTR]], [[VAL]], [[VAL]] offset1:8
12 define amdgpu_kernel void @simple_write2_one_val_f32(float addrspace(1)* %C, float addrspace(1)* %in) #0 {
13 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
14 %in.gep = getelementptr float, float addrspace(1)* %in, i32 %x.i
15 %val = load float, float addrspace(1)* %in.gep, align 4
16 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
17 store float %val, float addrspace(3)* %arrayidx0, align 4
18 %add.x = add nsw i32 %x.i, 8
19 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
20 store float %val, float addrspace(3)* %arrayidx1, align 4
24 ; SI-LABEL: @simple_write2_two_val_f32
25 ; SI-DAG: buffer_load_dword [[VAL0:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
26 ; SI-DAG: buffer_load_dword [[VAL1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
27 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 2, v{{[0-9]+}}
28 ; SI: ds_write2_b32 [[VPTR]], [[VAL0]], [[VAL1]] offset1:8
30 define amdgpu_kernel void @simple_write2_two_val_f32(float addrspace(1)* %C, float addrspace(1)* %in) #0 {
31 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
32 %in.gep.0 = getelementptr float, float addrspace(1)* %in, i32 %x.i
33 %in.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
34 %val0 = load volatile float, float addrspace(1)* %in.gep.0, align 4
35 %val1 = load volatile float, float addrspace(1)* %in.gep.1, align 4
36 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
37 store float %val0, float addrspace(3)* %arrayidx0, align 4
38 %add.x = add nsw i32 %x.i, 8
39 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
40 store float %val1, float addrspace(3)* %arrayidx1, align 4
44 ; SI-LABEL: @simple_write2_two_val_f32_volatile_0
45 ; SI-NOT: ds_write2_b32
46 ; SI: ds_write_b32 {{v[0-9]+}}, {{v[0-9]+}}
47 ; SI: ds_write_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:32
49 define amdgpu_kernel void @simple_write2_two_val_f32_volatile_0(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
50 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
51 %in0.gep = getelementptr float, float addrspace(1)* %in0, i32 %x.i
52 %in1.gep = getelementptr float, float addrspace(1)* %in1, i32 %x.i
53 %val0 = load volatile float, float addrspace(1)* %in0.gep, align 4
54 %val1 = load volatile float, float addrspace(1)* %in1.gep, align 4
55 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
56 store volatile float %val0, float addrspace(3)* %arrayidx0, align 4
57 %add.x = add nsw i32 %x.i, 8
58 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
59 store float %val1, float addrspace(3)* %arrayidx1, align 4
63 ; SI-LABEL: @simple_write2_two_val_f32_volatile_1
64 ; SI-NOT: ds_write2_b32
65 ; SI: ds_write_b32 {{v[0-9]+}}, {{v[0-9]+}}
66 ; SI: ds_write_b32 {{v[0-9]+}}, {{v[0-9]+}} offset:32
68 define amdgpu_kernel void @simple_write2_two_val_f32_volatile_1(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
69 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
70 %in0.gep = getelementptr float, float addrspace(1)* %in0, i32 %x.i
71 %in1.gep = getelementptr float, float addrspace(1)* %in1, i32 %x.i
72 %val0 = load volatile float, float addrspace(1)* %in0.gep, align 4
73 %val1 = load volatile float, float addrspace(1)* %in1.gep, align 4
74 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
75 store float %val0, float addrspace(3)* %arrayidx0, align 4
76 %add.x = add nsw i32 %x.i, 8
77 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
78 store volatile float %val1, float addrspace(3)* %arrayidx1, align 4
82 ; 2 data subregisters from different super registers.
83 ; SI-LABEL: @simple_write2_two_val_subreg2_mixed_f32
84 ; SI: buffer_load_dwordx2 v{{\[}}[[VAL0:[0-9]+]]:{{[0-9]+\]}}
85 ; SI: buffer_load_dwordx2 v{{\[[0-9]+}}:[[VAL1:[0-9]+]]{{\]}}
86 ; SI: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 2, v{{[0-9]+}}
87 ; SI: ds_write2_b32 [[VPTR]], v[[VAL0]], v[[VAL1]] offset1:8
89 define amdgpu_kernel void @simple_write2_two_val_subreg2_mixed_f32(float addrspace(1)* %C, <2 x float> addrspace(1)* %in) #0 {
90 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
91 %in.gep.0 = getelementptr <2 x float>, <2 x float> addrspace(1)* %in, i32 %x.i
92 %in.gep.1 = getelementptr <2 x float>, <2 x float> addrspace(1)* %in.gep.0, i32 1
93 %val0 = load volatile <2 x float>, <2 x float> addrspace(1)* %in.gep.0, align 8
94 %val1 = load volatile <2 x float>, <2 x float> addrspace(1)* %in.gep.1, align 8
95 %val0.0 = extractelement <2 x float> %val0, i32 0
96 %val1.1 = extractelement <2 x float> %val1, i32 1
97 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
98 store float %val0.0, float addrspace(3)* %arrayidx0, align 4
99 %add.x = add nsw i32 %x.i, 8
100 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
101 store float %val1.1, float addrspace(3)* %arrayidx1, align 4
105 ; SI-LABEL: @simple_write2_two_val_subreg2_f32
106 ; SI-DAG: buffer_load_dwordx2 v{{\[}}[[VAL0:[0-9]+]]:[[VAL1:[0-9]+]]{{\]}}
107 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 2, v{{[0-9]+}}
108 ; SI: ds_write2_b32 [[VPTR]], v[[VAL0]], v[[VAL1]] offset1:8
110 define amdgpu_kernel void @simple_write2_two_val_subreg2_f32(float addrspace(1)* %C, <2 x float> addrspace(1)* %in) #0 {
111 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
112 %in.gep = getelementptr <2 x float>, <2 x float> addrspace(1)* %in, i32 %x.i
113 %val = load <2 x float>, <2 x float> addrspace(1)* %in.gep, align 8
114 %val0 = extractelement <2 x float> %val, i32 0
115 %val1 = extractelement <2 x float> %val, i32 1
116 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
117 store float %val0, float addrspace(3)* %arrayidx0, align 4
118 %add.x = add nsw i32 %x.i, 8
119 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
120 store float %val1, float addrspace(3)* %arrayidx1, align 4
124 ; SI-LABEL: @simple_write2_two_val_subreg4_f32
125 ; SI-DAG: buffer_load_dwordx4 v{{\[}}[[VAL0:[0-9]+]]:[[VAL1:[0-9]+]]{{\]}}
126 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 2, v{{[0-9]+}}
127 ; SI: ds_write2_b32 [[VPTR]], v[[VAL0]], v[[VAL1]] offset1:8
129 define amdgpu_kernel void @simple_write2_two_val_subreg4_f32(float addrspace(1)* %C, <4 x float> addrspace(1)* %in) #0 {
130 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
131 %in.gep = getelementptr <4 x float>, <4 x float> addrspace(1)* %in, i32 %x.i
132 %val = load <4 x float>, <4 x float> addrspace(1)* %in.gep, align 16
133 %val0 = extractelement <4 x float> %val, i32 0
134 %val1 = extractelement <4 x float> %val, i32 3
135 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
136 store float %val0, float addrspace(3)* %arrayidx0, align 4
137 %add.x = add nsw i32 %x.i, 8
138 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
139 store float %val1, float addrspace(3)* %arrayidx1, align 4
143 ; SI-LABEL: @simple_write2_two_val_max_offset_f32
144 ; SI-DAG: buffer_load_dword [[VAL0:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
145 ; SI-DAG: buffer_load_dword [[VAL1:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4
146 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 2, v{{[0-9]+}}
147 ; SI: ds_write2_b32 [[VPTR]], [[VAL0]], [[VAL1]] offset1:255
149 define amdgpu_kernel void @simple_write2_two_val_max_offset_f32(float addrspace(1)* %C, float addrspace(1)* %in) #0 {
150 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
151 %in.gep.0 = getelementptr float, float addrspace(1)* %in, i32 %x.i
152 %in.gep.1 = getelementptr float, float addrspace(1)* %in.gep.0, i32 1
153 %val0 = load volatile float, float addrspace(1)* %in.gep.0, align 4
154 %val1 = load volatile float, float addrspace(1)* %in.gep.1, align 4
155 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
156 store float %val0, float addrspace(3)* %arrayidx0, align 4
157 %add.x = add nsw i32 %x.i, 255
158 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
159 store float %val1, float addrspace(3)* %arrayidx1, align 4
163 ; SI-LABEL: @simple_write2_two_val_too_far_f32
164 ; SI: ds_write_b32 v{{[0-9]+}}, v{{[0-9]+}}
165 ; SI: ds_write_b32 v{{[0-9]+}}, v{{[0-9]+}} offset:1028
167 define amdgpu_kernel void @simple_write2_two_val_too_far_f32(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
168 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
169 %in0.gep = getelementptr float, float addrspace(1)* %in0, i32 %x.i
170 %in1.gep = getelementptr float, float addrspace(1)* %in1, i32 %x.i
171 %val0 = load float, float addrspace(1)* %in0.gep, align 4
172 %val1 = load float, float addrspace(1)* %in1.gep, align 4
173 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %x.i
174 store float %val0, float addrspace(3)* %arrayidx0, align 4
175 %add.x = add nsw i32 %x.i, 257
176 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %add.x
177 store float %val1, float addrspace(3)* %arrayidx1, align 4
181 ; SI-LABEL: @simple_write2_two_val_f32_x2
182 ; SI: ds_write2_b32 [[BASEADDR:v[0-9]+]], [[VAL0:v[0-9]+]], [[VAL1:v[0-9]+]] offset1:8
183 ; SI: ds_write2_b32 [[BASEADDR:v[0-9]+]], [[VAL0]], [[VAL1]] offset0:11 offset1:27
185 define amdgpu_kernel void @simple_write2_two_val_f32_x2(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
186 %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
187 %in0.gep = getelementptr float, float addrspace(1)* %in0, i32 %tid.x
188 %in1.gep = getelementptr float, float addrspace(1)* %in1, i32 %tid.x
189 %val0 = load float, float addrspace(1)* %in0.gep, align 4
190 %val1 = load float, float addrspace(1)* %in1.gep, align 4
192 %idx.0 = add nsw i32 %tid.x, 0
193 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.0
194 store float %val0, float addrspace(3)* %arrayidx0, align 4
196 %idx.1 = add nsw i32 %tid.x, 8
197 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.1
198 store float %val1, float addrspace(3)* %arrayidx1, align 4
200 %idx.2 = add nsw i32 %tid.x, 11
201 %arrayidx2 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.2
202 store float %val0, float addrspace(3)* %arrayidx2, align 4
204 %idx.3 = add nsw i32 %tid.x, 27
205 %arrayidx3 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.3
206 store float %val1, float addrspace(3)* %arrayidx3, align 4
211 ; SI-LABEL: @simple_write2_two_val_f32_x2_nonzero_base
212 ; SI: ds_write2_b32 [[BASEADDR:v[0-9]+]], [[VAL0:v[0-9]+]], [[VAL1:v[0-9]+]] offset0:3 offset1:8
213 ; SI: ds_write2_b32 [[BASEADDR:v[0-9]+]], [[VAL0]], [[VAL1]] offset0:11 offset1:27
215 define amdgpu_kernel void @simple_write2_two_val_f32_x2_nonzero_base(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1) #0 {
216 %tid.x = tail call i32 @llvm.amdgcn.workitem.id.x() #1
217 %in0.gep = getelementptr float, float addrspace(1)* %in0, i32 %tid.x
218 %in1.gep = getelementptr float, float addrspace(1)* %in1, i32 %tid.x
219 %val0 = load float, float addrspace(1)* %in0.gep, align 4
220 %val1 = load float, float addrspace(1)* %in1.gep, align 4
222 %idx.0 = add nsw i32 %tid.x, 3
223 %arrayidx0 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.0
224 store float %val0, float addrspace(3)* %arrayidx0, align 4
226 %idx.1 = add nsw i32 %tid.x, 8
227 %arrayidx1 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.1
228 store float %val1, float addrspace(3)* %arrayidx1, align 4
230 %idx.2 = add nsw i32 %tid.x, 11
231 %arrayidx2 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.2
232 store float %val0, float addrspace(3)* %arrayidx2, align 4
234 %idx.3 = add nsw i32 %tid.x, 27
235 %arrayidx3 = getelementptr inbounds [512 x float], [512 x float] addrspace(3)* @lds, i32 0, i32 %idx.3
236 store float %val1, float addrspace(3)* %arrayidx3, align 4
241 ; SI-LABEL: @write2_ptr_subreg_arg_two_val_f32
242 ; SI-NOT: ds_write2_b32
246 define amdgpu_kernel void @write2_ptr_subreg_arg_two_val_f32(float addrspace(1)* %C, float addrspace(1)* %in0, float addrspace(1)* %in1, <2 x float addrspace(3)*> %lds.ptr) #0 {
247 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
248 %in0.gep = getelementptr float, float addrspace(1)* %in0, i32 %x.i
249 %in1.gep = getelementptr float, float addrspace(1)* %in1, i32 %x.i
250 %val0 = load float, float addrspace(1)* %in0.gep, align 4
251 %val1 = load float, float addrspace(1)* %in1.gep, align 4
253 %index.0 = insertelement <2 x i32> undef, i32 %x.i, i32 0
254 %index.1 = insertelement <2 x i32> %index.0, i32 8, i32 0
255 %gep = getelementptr inbounds float, <2 x float addrspace(3)*> %lds.ptr, <2 x i32> %index.1
256 %gep.0 = extractelement <2 x float addrspace(3)*> %gep, i32 0
257 %gep.1 = extractelement <2 x float addrspace(3)*> %gep, i32 1
259 ; Apply an additional offset after the vector that will be more obviously folded.
260 %gep.1.offset = getelementptr float, float addrspace(3)* %gep.1, i32 8
261 store float %val0, float addrspace(3)* %gep.0, align 4
263 %add.x = add nsw i32 %x.i, 8
264 store float %val1, float addrspace(3)* %gep.1.offset, align 4
268 ; SI-LABEL: @simple_write2_one_val_f64
269 ; SI-DAG: buffer_load_dwordx2 [[VAL:v\[[0-9]+:[0-9]+\]]],
270 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 3, v{{[0-9]+}}
271 ; SI: ds_write2_b64 [[VPTR]], [[VAL]], [[VAL]] offset1:8
273 define amdgpu_kernel void @simple_write2_one_val_f64(double addrspace(1)* %C, double addrspace(1)* %in) #0 {
274 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
275 %in.gep = getelementptr double, double addrspace(1)* %in, i32 %x.i
276 %val = load double, double addrspace(1)* %in.gep, align 8
277 %arrayidx0 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %x.i
278 store double %val, double addrspace(3)* %arrayidx0, align 8
279 %add.x = add nsw i32 %x.i, 8
280 %arrayidx1 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %add.x
281 store double %val, double addrspace(3)* %arrayidx1, align 8
285 ; SI-LABEL: @misaligned_simple_write2_one_val_f64
286 ; SI-DAG: buffer_load_dwordx2 v{{\[}}[[VAL0:[0-9]+]]:[[VAL1:[0-9]+]]{{\]}}
287 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 3, v{{[0-9]+}}
288 ; SI: ds_write2_b32 [[VPTR]], v[[VAL0]], v[[VAL1]] offset1:1
289 ; SI: ds_write2_b32 [[VPTR]], v[[VAL0]], v[[VAL1]] offset0:14 offset1:15
291 define amdgpu_kernel void @misaligned_simple_write2_one_val_f64(double addrspace(1)* %C, double addrspace(1)* %in, double addrspace(3)* %lds) #0 {
292 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
293 %in.gep = getelementptr double, double addrspace(1)* %in, i32 %x.i
294 %val = load double, double addrspace(1)* %in.gep, align 8
295 %arrayidx0 = getelementptr inbounds double, double addrspace(3)* %lds, i32 %x.i
296 store double %val, double addrspace(3)* %arrayidx0, align 4
297 %add.x = add nsw i32 %x.i, 7
298 %arrayidx1 = getelementptr inbounds double, double addrspace(3)* %lds, i32 %add.x
299 store double %val, double addrspace(3)* %arrayidx1, align 4
303 ; SI-LABEL: @simple_write2_two_val_f64
304 ; SI-DAG: buffer_load_dwordx2 [[VAL0:v\[[0-9]+:[0-9]+\]]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
305 ; SI-DAG: buffer_load_dwordx2 [[VAL1:v\[[0-9]+:[0-9]+\]]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:8
306 ; SI-DAG: v_lshlrev_b32_e32 [[VPTR:v[0-9]+]], 3, v{{[0-9]+}}
307 ; SI: ds_write2_b64 [[VPTR]], [[VAL0]], [[VAL1]] offset1:8
309 define amdgpu_kernel void @simple_write2_two_val_f64(double addrspace(1)* %C, double addrspace(1)* %in) #0 {
310 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
311 %in.gep.0 = getelementptr double, double addrspace(1)* %in, i32 %x.i
312 %in.gep.1 = getelementptr double, double addrspace(1)* %in.gep.0, i32 1
313 %val0 = load volatile double, double addrspace(1)* %in.gep.0, align 8
314 %val1 = load volatile double, double addrspace(1)* %in.gep.1, align 8
315 %arrayidx0 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %x.i
316 store double %val0, double addrspace(3)* %arrayidx0, align 8
317 %add.x = add nsw i32 %x.i, 8
318 %arrayidx1 = getelementptr inbounds [512 x double], [512 x double] addrspace(3)* @lds.f64, i32 0, i32 %add.x
319 store double %val1, double addrspace(3)* %arrayidx1, align 8
323 @foo = addrspace(3) global [4 x i32] undef, align 4
325 ; SI-LABEL: @store_constant_adjacent_offsets
326 ; SI: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0{{$}}
327 ; SI: ds_write2_b32 [[ZERO]], v{{[0-9]+}}, v{{[0-9]+}} offset1:1
328 define amdgpu_kernel void @store_constant_adjacent_offsets() {
329 store i32 123, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 0), align 4
330 store i32 123, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 1), align 4
334 ; SI-LABEL: @store_constant_disjoint_offsets
335 ; SI-DAG: v_mov_b32_e32 [[VAL:v[0-9]+]], 0x7b{{$}}
336 ; SI-DAG: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0{{$}}
337 ; SI: ds_write2_b32 [[ZERO]], [[VAL]], [[VAL]] offset1:2
338 define amdgpu_kernel void @store_constant_disjoint_offsets() {
339 store i32 123, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 0), align 4
340 store i32 123, i32 addrspace(3)* getelementptr inbounds ([4 x i32], [4 x i32] addrspace(3)* @foo, i32 0, i32 2), align 4
344 @bar = addrspace(3) global [4 x i64] undef, align 4
346 ; SI-LABEL: @store_misaligned64_constant_offsets
347 ; SI: v_mov_b32_e32 [[ZERO:v[0-9]+]], 0{{$}}
348 ; SI-DAG: ds_write2_b32 [[ZERO]], v{{[0-9]+}}, v{{[0-9]+}} offset1:1
349 ; SI-DAG: ds_write2_b32 [[ZERO]], v{{[0-9]+}}, v{{[0-9]+}} offset0:2 offset1:3
351 define amdgpu_kernel void @store_misaligned64_constant_offsets() {
352 store i64 123, i64 addrspace(3)* getelementptr inbounds ([4 x i64], [4 x i64] addrspace(3)* @bar, i32 0, i32 0), align 4
353 store i64 123, i64 addrspace(3)* getelementptr inbounds ([4 x i64], [4 x i64] addrspace(3)* @bar, i32 0, i32 1), align 4
357 @bar.large = addrspace(3) global [4096 x i64] undef, align 4
359 ; SI-LABEL: @store_misaligned64_constant_large_offsets
360 ; SI-DAG: v_mov_b32_e32 [[BASE0:v[0-9]+]], 0x7ff8{{$}}
361 ; SI-DAG: v_mov_b32_e32 [[BASE1:v[0-9]+]], 0x4000{{$}}
362 ; SI-DAG: ds_write2_b32 [[BASE0]], v{{[0-9]+}}, v{{[0-9]+}} offset1:1
363 ; SI-DAG: ds_write2_b32 [[BASE1]], v{{[0-9]+}}, v{{[0-9]+}} offset1:1
365 define amdgpu_kernel void @store_misaligned64_constant_large_offsets() {
366 store i64 123, i64 addrspace(3)* getelementptr inbounds ([4096 x i64], [4096 x i64] addrspace(3)* @bar.large, i32 0, i32 2048), align 4
367 store i64 123, i64 addrspace(3)* getelementptr inbounds ([4096 x i64], [4096 x i64] addrspace(3)* @bar.large, i32 0, i32 4095), align 4
371 @sgemm.lA = internal unnamed_addr addrspace(3) global [264 x float] undef, align 4
372 @sgemm.lB = internal unnamed_addr addrspace(3) global [776 x float] undef, align 4
374 define amdgpu_kernel void @write2_sgemm_sequence(float addrspace(1)* %C, i32 %lda, i32 %ldb, float addrspace(1)* %in) #0 {
375 %x.i = tail call i32 @llvm.amdgcn.workgroup.id.x() #1
376 %y.i = tail call i32 @llvm.amdgcn.workitem.id.y() #1
377 %val = load float, float addrspace(1)* %in
378 %arrayidx44 = getelementptr inbounds [264 x float], [264 x float] addrspace(3)* @sgemm.lA, i32 0, i32 %x.i
379 store float %val, float addrspace(3)* %arrayidx44, align 4
380 %add47 = add nsw i32 %x.i, 1
381 %arrayidx48 = getelementptr inbounds [264 x float], [264 x float] addrspace(3)* @sgemm.lA, i32 0, i32 %add47
382 store float %val, float addrspace(3)* %arrayidx48, align 4
383 %add51 = add nsw i32 %x.i, 16
384 %arrayidx52 = getelementptr inbounds [264 x float], [264 x float] addrspace(3)* @sgemm.lA, i32 0, i32 %add51
385 store float %val, float addrspace(3)* %arrayidx52, align 4
386 %add55 = add nsw i32 %x.i, 17
387 %arrayidx56 = getelementptr inbounds [264 x float], [264 x float] addrspace(3)* @sgemm.lA, i32 0, i32 %add55
388 store float %val, float addrspace(3)* %arrayidx56, align 4
389 %arrayidx60 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %y.i
390 store float %val, float addrspace(3)* %arrayidx60, align 4
391 %add63 = add nsw i32 %y.i, 1
392 %arrayidx64 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %add63
393 store float %val, float addrspace(3)* %arrayidx64, align 4
394 %add67 = add nsw i32 %y.i, 32
395 %arrayidx68 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %add67
396 store float %val, float addrspace(3)* %arrayidx68, align 4
397 %add71 = add nsw i32 %y.i, 33
398 %arrayidx72 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %add71
399 store float %val, float addrspace(3)* %arrayidx72, align 4
400 %add75 = add nsw i32 %y.i, 64
401 %arrayidx76 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %add75
402 store float %val, float addrspace(3)* %arrayidx76, align 4
403 %add79 = add nsw i32 %y.i, 65
404 %arrayidx80 = getelementptr inbounds [776 x float], [776 x float] addrspace(3)* @sgemm.lB, i32 0, i32 %add79
405 store float %val, float addrspace(3)* %arrayidx80, align 4
409 ; CI-LABEL: {{^}}simple_write2_v4f32_superreg_align4:
410 ; CI: ds_write2_b32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} offset0:3 offset1:2{{$}}
411 ; CI: ds_write2_b32 {{v[0-9]+}}, {{v[0-9]+}}, {{v[0-9]+}} offset0:1{{$}}
413 define amdgpu_kernel void @simple_write2_v4f32_superreg_align4(<4 x float> addrspace(3)* %out, <4 x float> addrspace(1)* %in) #0 {
414 %x.i = tail call i32 @llvm.amdgcn.workitem.id.x() #1
415 %in.gep = getelementptr inbounds <4 x float>, <4 x float> addrspace(1)* %in
416 %val0 = load <4 x float>, <4 x float> addrspace(1)* %in.gep, align 4
417 %out.gep = getelementptr inbounds <4 x float>, <4 x float> addrspace(3)* %out, i32 %x.i
418 store <4 x float> %val0, <4 x float> addrspace(3)* %out.gep, align 4
422 ; Function Attrs: nounwind readnone
423 declare i32 @llvm.amdgcn.workgroup.id.x() #1
425 ; Function Attrs: nounwind readnone
426 declare i32 @llvm.amdgcn.workgroup.id.y() #1
428 ; Function Attrs: nounwind readnone
429 declare i32 @llvm.amdgcn.workitem.id.x() #1
431 ; Function Attrs: nounwind readnone
432 declare i32 @llvm.amdgcn.workitem.id.y() #1
434 attributes #0 = { nounwind }
435 attributes #1 = { nounwind readnone }
436 attributes #2 = { convergent nounwind }