1 ; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs< %s | FileCheck -check-prefix=SI %s
2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs< %s | FileCheck -check-prefix=SI %s
4 ; SI-LABEL: {{^}}unaligned_load_store_i16_local:
10 define void @unaligned_load_store_i16_local(i16 addrspace(3)* %p, i16 addrspace(3)* %r) nounwind {
11 %v = load i16, i16 addrspace(3)* %p, align 1
12 store i16 %v, i16 addrspace(3)* %r, align 1
16 ; SI-LABEL: {{^}}unaligned_load_store_i16_global:
17 ; SI: buffer_load_ubyte
18 ; SI: buffer_load_ubyte
19 ; SI: buffer_store_byte
20 ; SI: buffer_store_byte
22 define void @unaligned_load_store_i16_global(i16 addrspace(1)* %p, i16 addrspace(1)* %r) nounwind {
23 %v = load i16, i16 addrspace(1)* %p, align 1
24 store i16 %v, i16 addrspace(1)* %r, align 1
28 ; SI-LABEL: {{^}}unaligned_load_store_i32_local:
38 define void @unaligned_load_store_i32_local(i32 addrspace(3)* %p, i32 addrspace(3)* %r) nounwind {
39 %v = load i32, i32 addrspace(3)* %p, align 1
40 store i32 %v, i32 addrspace(3)* %r, align 1
44 ; SI-LABEL: {{^}}unaligned_load_store_i32_global:
45 ; SI: buffer_load_ubyte
46 ; SI: buffer_load_ubyte
47 ; SI: buffer_load_ubyte
48 ; SI: buffer_load_ubyte
49 ; SI: buffer_store_byte
50 ; SI: buffer_store_byte
51 ; SI: buffer_store_byte
52 ; SI: buffer_store_byte
53 define void @unaligned_load_store_i32_global(i32 addrspace(1)* %p, i32 addrspace(1)* %r) nounwind {
54 %v = load i32, i32 addrspace(1)* %p, align 1
55 store i32 %v, i32 addrspace(1)* %r, align 1
59 ; SI-LABEL: {{^}}unaligned_load_store_i64_local:
77 define void @unaligned_load_store_i64_local(i64 addrspace(3)* %p, i64 addrspace(3)* %r) {
78 %v = load i64, i64 addrspace(3)* %p, align 1
79 store i64 %v, i64 addrspace(3)* %r, align 1
83 ; SI-LABEL: {{^}}unaligned_load_store_i64_global:
84 ; SI: buffer_load_ubyte
85 ; SI: buffer_load_ubyte
86 ; SI: buffer_load_ubyte
87 ; SI: buffer_load_ubyte
88 ; SI: buffer_load_ubyte
89 ; SI: buffer_load_ubyte
90 ; SI: buffer_load_ubyte
91 ; SI: buffer_load_ubyte
92 ; SI: buffer_store_byte
93 ; SI: buffer_store_byte
94 ; SI: buffer_store_byte
95 ; SI: buffer_store_byte
96 ; SI: buffer_store_byte
97 ; SI: buffer_store_byte
98 ; SI: buffer_store_byte
99 ; SI: buffer_store_byte
100 define void @unaligned_load_store_i64_global(i64 addrspace(1)* %p, i64 addrspace(1)* %r) {
101 %v = load i64, i64 addrspace(1)* %p, align 1
102 store i64 %v, i64 addrspace(1)* %r, align 1
106 ; SI-LABEL: {{^}}unaligned_load_store_v4i32_local:
147 define void @unaligned_load_store_v4i32_local(<4 x i32> addrspace(3)* %p, <4 x i32> addrspace(3)* %r) nounwind {
148 %v = load <4 x i32>, <4 x i32> addrspace(3)* %p, align 1
149 store <4 x i32> %v, <4 x i32> addrspace(3)* %r, align 1
153 ; FIXME: We mark v4i32 as custom, so misaligned loads are never expanded.
154 ; FIXME-SI-LABEL: {{^}}unaligned_load_store_v4i32_global
155 ; FIXME-SI: buffer_load_ubyte
156 ; FIXME-SI: buffer_load_ubyte
157 ; FIXME-SI: buffer_load_ubyte
158 ; FIXME-SI: buffer_load_ubyte
159 ; FIXME-SI: buffer_load_ubyte
160 ; FIXME-SI: buffer_load_ubyte
161 ; FIXME-SI: buffer_load_ubyte
162 ; FIXME-SI: buffer_load_ubyte
163 ; FIXME-SI: buffer_load_ubyte
164 ; FIXME-SI: buffer_load_ubyte
165 ; FIXME-SI: buffer_load_ubyte
166 ; FIXME-SI: buffer_load_ubyte
167 ; FIXME-SI: buffer_load_ubyte
168 ; FIXME-SI: buffer_load_ubyte
169 ; FIXME-SI: buffer_load_ubyte
170 ; FIXME-SI: buffer_load_ubyte
171 define void @unaligned_load_store_v4i32_global(<4 x i32> addrspace(1)* %p, <4 x i32> addrspace(1)* %r) nounwind {
172 %v = load <4 x i32>, <4 x i32> addrspace(1)* %p, align 1
173 store <4 x i32> %v, <4 x i32> addrspace(1)* %r, align 1
177 ; SI-LABEL: {{^}}load_lds_i64_align_4:
180 define void @load_lds_i64_align_4(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
181 %val = load i64, i64 addrspace(3)* %in, align 4
182 store i64 %val, i64 addrspace(1)* %out, align 8
186 ; SI-LABEL: {{^}}load_lds_i64_align_4_with_offset
187 ; SI: ds_read2_b32 v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]}} offset0:8 offset1:9
189 define void @load_lds_i64_align_4_with_offset(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
190 %ptr = getelementptr i64, i64 addrspace(3)* %in, i32 4
191 %val = load i64, i64 addrspace(3)* %ptr, align 4
192 store i64 %val, i64 addrspace(1)* %out, align 8
196 ; SI-LABEL: {{^}}load_lds_i64_align_4_with_split_offset:
197 ; The tests for the case where the lo offset is 8-bits, but the hi offset is 9-bits
198 ; SI: ds_read2_b32 v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]}} offset1:1
200 define void @load_lds_i64_align_4_with_split_offset(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
201 %ptr = bitcast i64 addrspace(3)* %in to i32 addrspace(3)*
202 %ptr255 = getelementptr i32, i32 addrspace(3)* %ptr, i32 255
203 %ptri64 = bitcast i32 addrspace(3)* %ptr255 to i64 addrspace(3)*
204 %val = load i64, i64 addrspace(3)* %ptri64, align 4
205 store i64 %val, i64 addrspace(1)* %out, align 8
209 ; SI-LABEL: {{^}}load_lds_i64_align_1:
218 ; SI: buffer_store_dwordx2
221 define void @load_lds_i64_align_1(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
222 %val = load i64, i64 addrspace(3)* %in, align 1
223 store i64 %val, i64 addrspace(1)* %out, align 8
227 ; SI-LABEL: {{^}}store_lds_i64_align_4:
230 define void @store_lds_i64_align_4(i64 addrspace(3)* %out, i64 %val) #0 {
231 store i64 %val, i64 addrspace(3)* %out, align 4
235 ; SI-LABEL: {{^}}store_lds_i64_align_4_with_offset
236 ; SI: ds_write2_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset0:8 offset1:9
238 define void @store_lds_i64_align_4_with_offset(i64 addrspace(3)* %out) #0 {
239 %ptr = getelementptr i64, i64 addrspace(3)* %out, i32 4
240 store i64 0, i64 addrspace(3)* %ptr, align 4
244 ; SI-LABEL: {{^}}store_lds_i64_align_4_with_split_offset:
245 ; The tests for the case where the lo offset is 8-bits, but the hi offset is 9-bits
246 ; SI: ds_write2_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset1:1
248 define void @store_lds_i64_align_4_with_split_offset(i64 addrspace(3)* %out) #0 {
249 %ptr = bitcast i64 addrspace(3)* %out to i32 addrspace(3)*
250 %ptr255 = getelementptr i32, i32 addrspace(3)* %ptr, i32 255
251 %ptri64 = bitcast i32 addrspace(3)* %ptr255 to i64 addrspace(3)*
252 store i64 0, i64 addrspace(3)* %out, align 4