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_i32_local:
11 define void @unaligned_load_store_i32_local(i32 addrspace(3)* %p, i32 addrspace(3)* %r) nounwind {
12 %v = load i32 addrspace(3)* %p, align 1
13 store i32 %v, i32 addrspace(3)* %r, align 1
17 ; SI-LABEL: {{^}}unaligned_load_store_i32_global:
18 ; SI: buffer_load_ubyte
19 ; SI: buffer_load_ubyte
20 ; SI: buffer_load_ubyte
21 ; SI: buffer_load_ubyte
22 ; SI: buffer_store_dword
23 define void @unaligned_load_store_i32_global(i32 addrspace(1)* %p, i32 addrspace(1)* %r) nounwind {
24 %v = load i32 addrspace(1)* %p, align 1
25 store i32 %v, i32 addrspace(1)* %r, align 1
29 ; SI-LABEL: {{^}}unaligned_load_store_i64_local:
41 define void @unaligned_load_store_i64_local(i64 addrspace(3)* %p, i64 addrspace(3)* %r) {
42 %v = load i64 addrspace(3)* %p, align 1
43 store i64 %v, i64 addrspace(3)* %r, align 1
47 ; SI-LABEL: {{^}}unaligned_load_store_i64_global:
48 ; SI: buffer_load_ubyte
49 ; SI: buffer_load_ubyte
50 ; SI: buffer_load_ubyte
51 ; SI: buffer_load_ubyte
52 ; SI: buffer_load_ubyte
53 ; SI: buffer_load_ubyte
54 ; SI: buffer_load_ubyte
55 ; SI: buffer_load_ubyte
56 ; SI: buffer_store_dword
57 ; SI: buffer_store_dword
58 define void @unaligned_load_store_i64_global(i64 addrspace(1)* %p, i64 addrspace(1)* %r) {
59 %v = load i64 addrspace(1)* %p, align 1
60 store i64 %v, i64 addrspace(1)* %r, align 1
64 ; SI-LABEL: {{^}}unaligned_load_store_v4i32_local:
90 define void @unaligned_load_store_v4i32_local(<4 x i32> addrspace(3)* %p, <4 x i32> addrspace(3)* %r) nounwind {
91 %v = load <4 x i32> addrspace(3)* %p, align 1
92 store <4 x i32> %v, <4 x i32> addrspace(3)* %r, align 1
96 ; FIXME: We mark v4i32 as custom, so misaligned loads are never expanded.
97 ; FIXME-SI-LABEL: {{^}}unaligned_load_store_v4i32_global
98 ; FIXME-SI: buffer_load_ubyte
99 ; FIXME-SI: buffer_load_ubyte
100 ; FIXME-SI: buffer_load_ubyte
101 ; FIXME-SI: buffer_load_ubyte
102 ; FIXME-SI: buffer_load_ubyte
103 ; FIXME-SI: buffer_load_ubyte
104 ; FIXME-SI: buffer_load_ubyte
105 ; FIXME-SI: buffer_load_ubyte
106 ; FIXME-SI: buffer_load_ubyte
107 ; FIXME-SI: buffer_load_ubyte
108 ; FIXME-SI: buffer_load_ubyte
109 ; FIXME-SI: buffer_load_ubyte
110 ; FIXME-SI: buffer_load_ubyte
111 ; FIXME-SI: buffer_load_ubyte
112 ; FIXME-SI: buffer_load_ubyte
113 ; FIXME-SI: buffer_load_ubyte
114 define void @unaligned_load_store_v4i32_global(<4 x i32> addrspace(1)* %p, <4 x i32> addrspace(1)* %r) nounwind {
115 %v = load <4 x i32> addrspace(1)* %p, align 1
116 store <4 x i32> %v, <4 x i32> addrspace(1)* %r, align 1
120 ; SI-LABEL: {{^}}load_lds_i64_align_4:
123 define void @load_lds_i64_align_4(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
124 %val = load i64 addrspace(3)* %in, align 4
125 store i64 %val, i64 addrspace(1)* %out, align 8
129 ; SI-LABEL: {{^}}load_lds_i64_align_4_with_offset
130 ; SI: ds_read2_b32 v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]}} offset0:8 offset1:9
132 define void @load_lds_i64_align_4_with_offset(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
133 %ptr = getelementptr i64 addrspace(3)* %in, i32 4
134 %val = load i64 addrspace(3)* %ptr, align 4
135 store i64 %val, i64 addrspace(1)* %out, align 8
139 ; SI-LABEL: {{^}}load_lds_i64_align_4_with_split_offset:
140 ; The tests for the case where the lo offset is 8-bits, but the hi offset is 9-bits
141 ; SI: ds_read2_b32 v[{{[0-9]+}}:{{[0-9]+}}], v{{[0-9]}} offset0:0 offset1:1
143 define void @load_lds_i64_align_4_with_split_offset(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
144 %ptr = bitcast i64 addrspace(3)* %in to i32 addrspace(3)*
145 %ptr255 = getelementptr i32 addrspace(3)* %ptr, i32 255
146 %ptri64 = bitcast i32 addrspace(3)* %ptr255 to i64 addrspace(3)*
147 %val = load i64 addrspace(3)* %ptri64, align 4
148 store i64 %val, i64 addrspace(1)* %out, align 8
152 ; SI: {{^}}load_lds_i64_align_1
162 define void @load_lds_i64_align_1(i64 addrspace(1)* nocapture %out, i64 addrspace(3)* %in) #0 {
163 %val = load i64 addrspace(3)* %in, align 1
164 store i64 %val, i64 addrspace(1)* %out, align 8
168 ; SI-LABEL: {{^}}store_lds_i64_align_4:
171 define void @store_lds_i64_align_4(i64 addrspace(3)* %out, i64 %val) #0 {
172 store i64 %val, i64 addrspace(3)* %out, align 4
176 ; SI-LABEL: {{^}}store_lds_i64_align_4_with_offset
177 ; SI: ds_write2_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset0:8 offset1:9
179 define void @store_lds_i64_align_4_with_offset(i64 addrspace(3)* %out) #0 {
180 %ptr = getelementptr i64 addrspace(3)* %out, i32 4
181 store i64 0, i64 addrspace(3)* %ptr, align 4
185 ; SI-LABEL: {{^}}store_lds_i64_align_4_with_split_offset:
186 ; The tests for the case where the lo offset is 8-bits, but the hi offset is 9-bits
187 ; SI: ds_write2_b32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} offset0:0 offset1:1
189 define void @store_lds_i64_align_4_with_split_offset(i64 addrspace(3)* %out) #0 {
190 %ptr = bitcast i64 addrspace(3)* %out to i32 addrspace(3)*
191 %ptr255 = getelementptr i32 addrspace(3)* %ptr, i32 255
192 %ptri64 = bitcast i32 addrspace(3)* %ptr255 to i64 addrspace(3)*
193 store i64 0, i64 addrspace(3)* %out, align 4